├── Code_Baddy ├── LoadBatches.py ├── LoadBatches.pyc ├── Models │ ├── TrackNet.py │ └── __pycache__ │ │ ├── TrackNet.cpython-37.pyc │ │ ├── __init__.cpython-35.pyc │ │ └── __init__.cpython-37.pyc ├── predict.py ├── predict_video.py ├── train.py └── weights │ └── model_baddy.h5 ├── Code_Tennis ├── LoadBatches.py ├── LoadBatches.pyc ├── Models │ ├── TrackNet.py │ └── __pycache__ │ │ ├── TrackNet.cpython-37.pyc │ │ ├── __init__.cpython-35.pyc │ │ └── __init__.cpython-37.pyc ├── predict.py ├── predict_video.py ├── train.py └── weights │ └── model_tennis.h5 ├── LICENSE ├── LabelingTool ├── LabelingTool.fig ├── LabelingTool.m ├── LabelingTool.m~ ├── LabelingTool.prj ├── LabelingTool │ ├── PackagingLog.html │ ├── Video_to_Frame.py │ ├── for_redistribution │ │ └── MyAppInstaller_web.exe │ ├── for_redistribution_files_only │ │ ├── LabelingTool.exe │ │ ├── default_icon.ico │ │ ├── readme.txt │ │ └── splash.png │ └── for_testing │ │ ├── LabelingTool.exe │ │ ├── mccExcludedFiles.log │ │ ├── readme.txt │ │ ├── requiredMCRProducts.txt │ │ └── splash.png ├── README.md └── t.ps ├── README.md ├── Resources ├── README.md ├── TrackNet.pdf ├── TrackNet.png └── coachAI.pdf ├── TrackNet.gif ├── TrackNet_dataset.ipynb ├── Utils └── frame_to_vid.py └── requirements.txt /Code_Baddy/LoadBatches.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import cv2 3 | import itertools 4 | import csv 5 | from collections import defaultdict 6 | 7 | 8 | #get input array 9 | def getInputArr( path ,path1 ,path2 , width , height): 10 | try: 11 | #read the image 12 | img = cv2.imread(path, 1) 13 | #resize it 14 | img = cv2.resize(img, ( width , height )) 15 | #input must be float type 16 | img = img.astype(np.float32) 17 | 18 | #read the image 19 | img1 = cv2.imread(path1, 1) 20 | #resize it 21 | img1 = cv2.resize(img1, ( width , height )) 22 | #input must be float type 23 | img1 = img1.astype(np.float32) 24 | 25 | #read the image 26 | img2 = cv2.imread(path2, 1) 27 | #resize it 28 | img2 = cv2.resize(img2, ( width , height )) 29 | #input must be float type 30 | img2 = img2.astype(np.float32) 31 | 32 | #combine three imgs to (width , height, rgb*3) 33 | imgs = np.concatenate((img, img1, img2),axis=2) 34 | 35 | #since the odering of TrackNet is 'channels_first', so we need to change the axis 36 | imgs = np.rollaxis(imgs, 2, 0) 37 | return imgs 38 | 39 | except Exception as e: 40 | 41 | print(path, e) 42 | 43 | 44 | 45 | #get output array 46 | def getOutputArr( path , nClasses , width , height ): 47 | 48 | seg_labels = np.zeros(( height , width , nClasses )) 49 | try: 50 | img = cv2.imread(path, 1) 51 | img = cv2.resize(img, ( width , height )) 52 | img = img[:, : , 0] 53 | np.savetxt("tracknet_img.txt", img, newline="\n") 54 | 55 | for c in range(nClasses): 56 | seg_labels[: , : , c ] = (img == c ).astype(int) 57 | 58 | except Exception as e: 59 | print(e) 60 | 61 | seg_labels = np.reshape(seg_labels, ( width*height , nClasses )) 62 | return seg_labels 63 | 64 | 65 | 66 | #read input data and output data 67 | def InputOutputGenerator( images_path, batch_size, n_classes , input_height , input_width , output_height , output_width ): 68 | 69 | 70 | #read csv file to 'zipped' 71 | columns = defaultdict(list) 72 | with open(images_path) as f: 73 | reader = csv.reader(f) 74 | next(reader) 75 | for row in reader: 76 | for (i,v) in enumerate(row): 77 | columns[i].append(v) 78 | zipped = itertools.cycle( zip(columns[0], columns[1], columns[2], columns[3]) ) 79 | 80 | while True: 81 | Input = [] 82 | Output = [] 83 | #read input&output for each batch 84 | for _ in range( batch_size) : 85 | path, path1, path2 , anno = next(zipped) 86 | Input.append( getInputArr(path, path1, path2 , input_width , input_height) ) 87 | Output.append( getOutputArr( anno , n_classes , output_width , output_height) ) 88 | #return input&output 89 | yield np.array(Input) , np.array(Output) 90 | 91 | -------------------------------------------------------------------------------- /Code_Baddy/LoadBatches.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Baddy/LoadBatches.pyc -------------------------------------------------------------------------------- /Code_Baddy/Models/TrackNet.py: -------------------------------------------------------------------------------- 1 | from keras.models import * 2 | from keras.layers import * 3 | 4 | def TrackNet( n_classes , input_height, input_width ): # input_height = 360, input_width = 640 5 | 6 | imgs_input = Input(shape=(9,input_height,input_width)) 7 | 8 | #layer1 9 | x = Conv2D(64, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(imgs_input) 10 | x = ( Activation('relu'))(x) 11 | x = ( BatchNormalization())(x) 12 | 13 | #layer2 14 | x = Conv2D(64, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 15 | x = ( Activation('relu'))(x) 16 | x = ( BatchNormalization())(x) 17 | 18 | #layer3 19 | x = MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_first' )(x) 20 | 21 | #layer4 22 | x = Conv2D(128, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 23 | x = ( Activation('relu'))(x) 24 | x = ( BatchNormalization())(x) 25 | 26 | #layer5 27 | x = Conv2D(128, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 28 | x = ( Activation('relu'))(x) 29 | x = ( BatchNormalization())(x) 30 | 31 | #layer6 32 | x = MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_first' )(x) 33 | 34 | #layer7 35 | x = Conv2D(256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 36 | x = ( Activation('relu'))(x) 37 | x = ( BatchNormalization())(x) 38 | 39 | #layer8 40 | x = Conv2D(256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 41 | x = ( Activation('relu'))(x) 42 | x = ( BatchNormalization())(x) 43 | 44 | #layer9 45 | x = Conv2D(256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 46 | x = ( Activation('relu'))(x) 47 | x = ( BatchNormalization())(x) 48 | 49 | #layer10 50 | x = MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_first' )(x) 51 | 52 | #layer11 53 | x = ( Conv2D(512, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 54 | x = ( Activation('relu'))(x) 55 | x = ( BatchNormalization())(x) 56 | 57 | #layer12 58 | x = ( Conv2D(512, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 59 | x = ( Activation('relu'))(x) 60 | x = ( BatchNormalization())(x) 61 | 62 | #layer13 63 | x = ( Conv2D(512, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 64 | x = ( Activation('relu'))(x) 65 | x = ( BatchNormalization())(x) 66 | 67 | #layer14 68 | x = ( UpSampling2D( (2,2), data_format='channels_first'))(x) 69 | 70 | #layer15 71 | x = ( Conv2D( 256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 72 | x = ( Activation('relu'))(x) 73 | x = ( BatchNormalization())(x) 74 | 75 | #layer16 76 | x = ( Conv2D( 256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 77 | x = ( Activation('relu'))(x) 78 | x = ( BatchNormalization())(x) 79 | 80 | #layer17 81 | x = ( Conv2D( 256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 82 | x = ( Activation('relu'))(x) 83 | x = ( BatchNormalization())(x) 84 | 85 | #layer18 86 | x = ( UpSampling2D( (2,2), data_format='channels_first'))(x) 87 | 88 | #layer19 89 | x = ( Conv2D( 128 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 90 | x = ( Activation('relu'))(x) 91 | x = ( BatchNormalization())(x) 92 | 93 | #layer20 94 | x = ( Conv2D( 128 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 95 | x = ( Activation('relu'))(x) 96 | x = ( BatchNormalization())(x) 97 | 98 | #layer21 99 | x = ( UpSampling2D( (2,2), data_format='channels_first'))(x) 100 | 101 | #layer22 102 | x = ( Conv2D( 64 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 103 | x = ( Activation('relu'))(x) 104 | x = ( BatchNormalization())(x) 105 | 106 | #layer23 107 | x = ( Conv2D( 64 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 108 | x = ( Activation('relu'))(x) 109 | x = ( BatchNormalization())(x) 110 | 111 | #layer24 112 | x = Conv2D( n_classes , (3, 3) , kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 113 | x = ( Activation('relu'))(x) 114 | x = ( BatchNormalization())(x) 115 | 116 | o_shape = Model(imgs_input , x ).output_shape 117 | print("layer24 output shape:", o_shape[1],o_shape[2],o_shape[3]) 118 | #layer24 output shape: 256, 360, 640 119 | 120 | OutputHeight = o_shape[2] 121 | OutputWidth = o_shape[3] 122 | 123 | #reshape the size to (256, 360*640) 124 | x = (Reshape(( -1 , OutputHeight*OutputWidth )))(x) 125 | 126 | #change dimension order to (360*640, 256) 127 | x = (Permute((2, 1)))(x) 128 | 129 | #layer25 130 | gaussian_output = (Activation('softmax'))(x) 131 | 132 | model = Model( imgs_input , gaussian_output) 133 | model.outputWidth = OutputWidth 134 | model.outputHeight = OutputHeight 135 | 136 | #show model's details 137 | model.summary() 138 | 139 | return model 140 | 141 | 142 | 143 | 144 | -------------------------------------------------------------------------------- /Code_Baddy/Models/__pycache__/TrackNet.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Baddy/Models/__pycache__/TrackNet.cpython-37.pyc -------------------------------------------------------------------------------- /Code_Baddy/Models/__pycache__/__init__.cpython-35.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Baddy/Models/__pycache__/__init__.cpython-35.pyc -------------------------------------------------------------------------------- /Code_Baddy/Models/__pycache__/__init__.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Baddy/Models/__pycache__/__init__.cpython-37.pyc -------------------------------------------------------------------------------- /Code_Baddy/predict.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import Models , LoadBatches 3 | import cv2 4 | import numpy as np 5 | import glob 6 | import os 7 | 8 | #parse parameters 9 | parser = argparse.ArgumentParser() 10 | parser.add_argument("--save_weights_path", type = str ) 11 | parser.add_argument("--test_images_path", type = str , default = "") 12 | parser.add_argument("--output_path", type = str , default = "") 13 | parser.add_argument("--input_height", type=int , default = 360 ) 14 | parser.add_argument("--input_width", type=int , default = 640 ) 15 | parser.add_argument("--output_height", type=int , default = 224 ) 16 | parser.add_argument("--output_width", type=int , default = 224 ) 17 | parser.add_argument("--n_classes", type=int ) 18 | 19 | args = parser.parse_args() 20 | n_classes = args.n_classes 21 | images_path = args.test_images_path 22 | output_path = args.output_path 23 | input_width = args.input_width 24 | input_height = args.input_height 25 | output_width = args.output_width 26 | output_height = args.output_height 27 | 28 | #load TrackNet model 29 | modelTN = Models.TrackNet.TrackNet 30 | m = modelTN( n_classes , input_height=input_height, input_width=input_width ) 31 | m.compile(loss='categorical_crossentropy', optimizer= 'adadelta', metrics=['accuracy']) 32 | m.load_weights( args.save_weights_path ) 33 | 34 | 35 | #get TrackNet output height and width 36 | model_output_height = m.outputHeight 37 | model_output_width = m.outputWidth 38 | 39 | #create grayscale RGB (0,0,0)~(255,255,255) 40 | colors = [ ( i, i, i ) for i in range(0, n_classes) ] 41 | 42 | #predict each clips from 1 to 82 43 | for clip in range(1,82): 44 | 45 | #get all JPG images in the path 46 | images = glob.glob( images_path + str(clip) + "/*.jpg" ) 47 | images.sort() 48 | 49 | 50 | #create folder for saving output image 51 | if not os.path.exists(output_path + str(clip) + "/"): 52 | os.makedirs(output_path + str(clip) + "/") 53 | 54 | #predict each images 55 | #since TrackNet cant predict first and second images, so we start from third image 56 | for i in range(2,len(images)): 57 | 58 | output_name = images[i].replace( images_path, output_path) 59 | 60 | #load input data 61 | X = LoadBatches.getInputArr( images[i], images[i-1], images[i-2], input_width, input_height ) 62 | #prdict heatmap 63 | pr = m.predict( np.array([X]) )[0] 64 | 65 | #since TrackNet output is ( net_output_height*model_output_width , n_classes ) 66 | #so we need to reshape image as ( net_output_height, model_output_width , n_classes(depth) ) 67 | pr = pr.reshape(( model_output_height, model_output_width , n_classes ) ).argmax( axis=2 ) 68 | 69 | #declare variable for output image 70 | output_img = np.zeros( ( model_output_height, model_output_width , 3 ) ) 71 | 72 | #set RGB value for each pixel predciton 73 | for c in range(n_classes): 74 | output_img[:,:,0] += ( (pr[:,: ] == c )*( colors[c][0] )).astype('uint8') 75 | output_img[:,:,1] += ((pr[:,: ] == c )*( colors[c][1] )).astype('uint8') 76 | output_img[:,:,2] += ((pr[:,: ] == c )*( colors[c][2] )).astype('uint8') 77 | 78 | #reshape the image size as original input image 79 | output_img = cv2.resize(output_img , (output_width , output_height )) 80 | 81 | #output heatmap image 82 | cv2.imwrite( output_name , output_img ) 83 | -------------------------------------------------------------------------------- /Code_Baddy/predict_video.py: -------------------------------------------------------------------------------- 1 | # python predict_video.py --save_weights_path=weights/model_baddy.h5 --input_video_path="badminton.mp4" --output_video_path="tracked.mp4" --n_classes=256 2 | 3 | import argparse 4 | # import Models 5 | from Models.TrackNet import TrackNet 6 | import queue 7 | import cv2 8 | import numpy as np 9 | from PIL import Image, ImageDraw 10 | 11 | #parse parameters 12 | parser = argparse.ArgumentParser() 13 | parser.add_argument("--input_video_path", type=str) 14 | parser.add_argument("--output_video_path", type=str, default = "") 15 | parser.add_argument("--save_weights_path", type = str ) 16 | parser.add_argument("--n_classes", type=int ) 17 | 18 | args = parser.parse_args() 19 | input_video_path = args.input_video_path 20 | output_video_path = args.output_video_path 21 | save_weights_path = args.save_weights_path 22 | n_classes = args.n_classes 23 | 24 | if output_video_path == "": 25 | #output video in same path 26 | output_video_path = input_video_path.split('.')[0] + "_TrackNet.mp4" 27 | 28 | #get video fps&video size 29 | video = cv2.VideoCapture(input_video_path) 30 | fps = int(video.get(cv2.CAP_PROP_FPS)) 31 | output_width = int(video.get(cv2.CAP_PROP_FRAME_WIDTH)) 32 | output_height = int(video.get(cv2.CAP_PROP_FRAME_HEIGHT)) 33 | 34 | #start from first frame 35 | currentFrame = 0 36 | 37 | #width and height in TrackNet 38 | width , height = 640, 360 39 | img, img1, img2 = None, None, None 40 | 41 | #load TrackNet model 42 | modelFN = TrackNet 43 | m = modelFN( n_classes , input_height=height, input_width=width ) 44 | m.compile(loss='categorical_crossentropy', optimizer= 'adadelta' , metrics=['accuracy']) 45 | m.load_weights( save_weights_path ) 46 | 47 | # In order to draw the trajectory of tennis, we need to save the coordinate of previous 7 frames 48 | q = queue.deque() 49 | for i in range(0,8): 50 | q.appendleft(None) 51 | 52 | #save prediction images as vidoe 53 | #Tutorial: https://stackoverflow.com/questions/33631489/error-during-saving-a-video-using-python-and-opencv 54 | fourcc = cv2.VideoWriter_fourcc(*'XVID') 55 | output_video = cv2.VideoWriter(output_video_path,fourcc, fps, (output_width,output_height)) 56 | 57 | 58 | #both first and second frames cant be predict, so we directly write the frames to output video 59 | #capture frame-by-frame 60 | video.set(1,currentFrame); 61 | ret, img1 = video.read() 62 | #write image to video 63 | output_video.write(img1) 64 | currentFrame +=1 65 | #resize it 66 | img1 = cv2.resize(img1, ( width , height )) 67 | #input must be float type 68 | img1 = img1.astype(np.float32) 69 | 70 | #capture frame-by-frame 71 | video.set(1,currentFrame); 72 | ret, img = video.read() 73 | #write image to video 74 | output_video.write(img) 75 | currentFrame +=1 76 | #resize it 77 | img = cv2.resize(img, ( width , height )) 78 | #input must be float type 79 | img = img.astype(np.float32) 80 | 81 | 82 | 83 | while(True): 84 | 85 | img2 = img1 86 | img1 = img 87 | 88 | #capture frame-by-frame 89 | video.set(1,currentFrame); 90 | ret, img = video.read() 91 | 92 | #if there dont have any frame in video, break 93 | if not ret: 94 | break 95 | 96 | #img is the frame that TrackNet will predict the position 97 | #since we need to change the size and type of img, copy it to output_img 98 | output_img = img 99 | 100 | #resize it 101 | img = cv2.resize(img, ( width , height )) 102 | #input must be float type 103 | img = img.astype(np.float32) 104 | 105 | 106 | #combine three imgs to (width , height, rgb*3) 107 | X = np.concatenate((img, img1, img2),axis=2) 108 | 109 | #since the odering of TrackNet is 'channels_first', so we need to change the axis 110 | X = np.rollaxis(X, 2, 0) 111 | #prdict heatmap 112 | pr = m.predict( np.array([X]) )[0] 113 | 114 | #since TrackNet output is ( net_output_height*model_output_width , n_classes ) 115 | #so we need to reshape image as ( net_output_height, model_output_width , n_classes(depth) ) 116 | #.argmax( axis=2 ) => select the largest probability as class 117 | pr = pr.reshape(( height , width , n_classes ) ).argmax( axis=2 ) 118 | 119 | #cv2 image must be numpy.uint8, convert numpy.int64 to numpy.uint8 120 | pr = pr.astype(np.uint8) 121 | 122 | #reshape the image size as original input image 123 | heatmap = cv2.resize(pr , (output_width, output_height )) 124 | 125 | #heatmap is converted into a binary image by threshold method. 126 | ret,heatmap = cv2.threshold(heatmap,127,255,cv2.THRESH_BINARY) 127 | 128 | #find the circle in image with 2<=radius<=7 129 | circles = cv2.HoughCircles(heatmap, cv2.HOUGH_GRADIENT,dp=1,minDist=1,param1=50,param2=2,minRadius=2,maxRadius=7) 130 | 131 | #In order to draw the circle in output_img, we need to used PIL library 132 | #Convert opencv image format to PIL image format 133 | PIL_image = cv2.cvtColor(output_img, cv2.COLOR_BGR2RGB) 134 | PIL_image = Image.fromarray(PIL_image) 135 | 136 | #check if there have any tennis be detected 137 | if circles is not None: 138 | #if only one tennis be detected 139 | if len(circles) == 1: 140 | 141 | x = int(circles[0][0][0]) 142 | y = int(circles[0][0][1]) 143 | print(currentFrame, x,y) 144 | 145 | #push x,y to queue 146 | q.appendleft([x,y]) 147 | #pop x,y from queue 148 | q.pop() 149 | else: 150 | #push None to queue 151 | q.appendleft(None) 152 | #pop x,y from queue 153 | q.pop() 154 | else: 155 | #push None to queue 156 | q.appendleft(None) 157 | #pop x,y from queue 158 | q.pop() 159 | 160 | #draw current frame prediction and previous 7 frames as yellow circle, total: 8 frames 161 | for i in range(0,8): 162 | if q[i] is not None: 163 | draw_x = q[i][0] 164 | draw_y = q[i][1] 165 | bbox = (draw_x - 2, draw_y - 2, draw_x + 2, draw_y + 2) 166 | draw = ImageDraw.Draw(PIL_image) 167 | draw.ellipse(bbox, outline ='yellow') 168 | del draw 169 | 170 | #Convert PIL image format back to opencv image format 171 | opencvImage = cv2.cvtColor(np.array(PIL_image), cv2.COLOR_RGB2BGR) 172 | #write image to output_video 173 | output_video.write(opencvImage) 174 | 175 | #next frame 176 | currentFrame += 1 177 | 178 | # everything is done, release the video 179 | video.release() 180 | output_video.release() 181 | print("finish") -------------------------------------------------------------------------------- /Code_Baddy/train.py: -------------------------------------------------------------------------------- 1 | # python train.py --save_weights_path=weights/model --training_images_name="training.csv" --epochs=500 --n_classes=256 --input_height=360 --input_width=640 --load_weights=2 --step_per_epochs=200 --batch_size=2 2 | # python train.py --save_weights_path=weights/model --training_images_name="training_100.csv" --epochs=50 --n_classes=256 --input_height=360 --input_width=640 --load_weights=2 --step_per_epochs=5 --batch_size=1 3 | # python train.py --save_weights_path=weights/model --training_images_name="training_100.csv" --epochs=50 --n_classes=256 --input_height=180 --input_width=320 --load_weights=-1 --step_per_epochs=5 --batch_size=1 4 | import argparse 5 | import Models , LoadBatches 6 | from keras import optimizers 7 | from keras.utils import plot_model 8 | from Models.TrackNet import TrackNet 9 | 10 | #parse parameters 11 | parser = argparse.ArgumentParser() 12 | parser.add_argument("--save_weights_path", type = str ) 13 | parser.add_argument("--training_images_name", type = str ) 14 | parser.add_argument("--n_classes", type=int ) 15 | parser.add_argument("--input_height", type=int , default = 360 ) 16 | parser.add_argument("--input_width", type=int , default = 640 ) 17 | parser.add_argument("--epochs", type = int, default = 1000 ) 18 | parser.add_argument("--batch_size", type = int, default = 2 ) 19 | parser.add_argument("--load_weights", type = str , default = "-1") 20 | parser.add_argument("--step_per_epochs", type = int, default = 200 ) 21 | 22 | args = parser.parse_args() 23 | training_images_name = args.training_images_name 24 | train_batch_size = args.batch_size 25 | n_classes = args.n_classes 26 | input_height = args.input_height 27 | input_width = args.input_width 28 | save_weights_path = args.save_weights_path 29 | epochs = args.epochs 30 | load_weights = args.load_weights 31 | step_per_epochs = args.step_per_epochs 32 | optimizer_name = optimizers.Adadelta(lr=1.0) 33 | 34 | #load TrackNet model 35 | modelTN = Models.TrackNet.TrackNet 36 | m = modelTN( n_classes , input_height=input_height, input_width=input_width ) 37 | m.compile(loss='categorical_crossentropy', optimizer= optimizer_name, metrics=['accuracy']) 38 | 39 | #check if need to retrain the model weights 40 | if load_weights != "-1": 41 | m.load_weights("weights/model." + load_weights) 42 | 43 | #show TrackNet details, save it as TrackNet.png 44 | plot_model( m , show_shapes=True , to_file='TrackNet.png') 45 | 46 | #get TrackNet output height and width 47 | model_output_height = m.outputHeight 48 | model_output_width = m.outputWidth 49 | 50 | #creat input data and output data 51 | Generator = LoadBatches.InputOutputGenerator( training_images_name, train_batch_size, n_classes , input_height , input_width , model_output_height , model_output_width) 52 | 53 | 54 | #start to train the model, and save weights until finish 55 | ''' 56 | m.fit_generator( Generator, step_per_epochs, epochs ) 57 | m.save_weights( save_weights_path + ".0" ) 58 | 59 | ''' 60 | 61 | #start to train the model, and save weights per 50 epochs 62 | 63 | for ep in range(1, epochs+1 ): 64 | print("Epoch :", str(ep) + "/" + str(epochs)) 65 | m.fit_generator(Generator, step_per_epochs, use_multiprocessing=True) 66 | if ep % 2 == 0: 67 | m.save_weights(save_weights_path + ".0") 68 | -------------------------------------------------------------------------------- /Code_Baddy/weights/model_baddy.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Baddy/weights/model_baddy.h5 -------------------------------------------------------------------------------- /Code_Tennis/LoadBatches.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import cv2 3 | import itertools 4 | import csv 5 | from collections import defaultdict 6 | 7 | 8 | #get input array 9 | def getInputArr( path ,path1 ,path2 , width , height): 10 | try: 11 | #read the image 12 | img = cv2.imread(path, 1) 13 | #resize it 14 | img = cv2.resize(img, ( width , height )) 15 | #input must be float type 16 | img = img.astype(np.float32) 17 | 18 | #read the image 19 | img1 = cv2.imread(path1, 1) 20 | #resize it 21 | img1 = cv2.resize(img1, ( width , height )) 22 | #input must be float type 23 | img1 = img1.astype(np.float32) 24 | 25 | #read the image 26 | img2 = cv2.imread(path2, 1) 27 | #resize it 28 | img2 = cv2.resize(img2, ( width , height )) 29 | #input must be float type 30 | img2 = img2.astype(np.float32) 31 | 32 | #combine three imgs to (width , height, rgb*3) 33 | imgs = np.concatenate((img, img1, img2),axis=2) 34 | 35 | #since the odering of TrackNet is 'channels_first', so we need to change the axis 36 | imgs = np.rollaxis(imgs, 2, 0) 37 | return imgs 38 | 39 | except Exception, e: 40 | 41 | print path , e 42 | 43 | 44 | 45 | #get output array 46 | def getOutputArr( path , nClasses , width , height ): 47 | 48 | seg_labels = np.zeros(( height , width , nClasses )) 49 | try: 50 | img = cv2.imread(path, 1) 51 | img = cv2.resize(img, ( width , height )) 52 | img = img[:, : , 0] 53 | 54 | for c in range(nClasses): 55 | seg_labels[: , : , c ] = (img == c ).astype(int) 56 | 57 | except Exception, e: 58 | print e 59 | 60 | seg_labels = np.reshape(seg_labels, ( width*height , nClasses )) 61 | return seg_labels 62 | 63 | 64 | 65 | #read input data and output data 66 | def InputOutputGenerator( images_path, batch_size, n_classes , input_height , input_width , output_height , output_width ): 67 | 68 | 69 | #read csv file to 'zipped' 70 | columns = defaultdict(list) 71 | with open(images_path) as f: 72 | reader = csv.reader(f) 73 | reader.next() 74 | for row in reader: 75 | for (i,v) in enumerate(row): 76 | columns[i].append(v) 77 | zipped = itertools.cycle( zip(columns[0], columns[1], columns[2], columns[3]) ) 78 | 79 | while True: 80 | Input = [] 81 | Output = [] 82 | #read input&output for each batch 83 | for _ in range( batch_size) : 84 | path, path1, path2 , anno = zipped.next() 85 | Input.append( getInputArr(path, path1, path2 , input_width , input_height) ) 86 | Output.append( getOutputArr( anno , n_classes , output_width , output_height) ) 87 | #return input&output 88 | yield np.array(Input) , np.array(Output) 89 | 90 | -------------------------------------------------------------------------------- /Code_Tennis/LoadBatches.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Tennis/LoadBatches.pyc -------------------------------------------------------------------------------- /Code_Tennis/Models/TrackNet.py: -------------------------------------------------------------------------------- 1 | from keras.models import * 2 | from keras.layers import * 3 | 4 | def TrackNet( n_classes , input_height, input_width ): # input_height = 360, input_width = 640 5 | 6 | imgs_input = Input(shape=(9,input_height,input_width)) 7 | 8 | #layer1 9 | x = Conv2D(64, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(imgs_input) 10 | x = ( Activation('relu'))(x) 11 | x = ( BatchNormalization())(x) 12 | 13 | #layer2 14 | x = Conv2D(64, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 15 | x = ( Activation('relu'))(x) 16 | x = ( BatchNormalization())(x) 17 | 18 | #layer3 19 | x = MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_first' )(x) 20 | 21 | #layer4 22 | x = Conv2D(128, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 23 | x = ( Activation('relu'))(x) 24 | x = ( BatchNormalization())(x) 25 | 26 | #layer5 27 | x = Conv2D(128, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 28 | x = ( Activation('relu'))(x) 29 | x = ( BatchNormalization())(x) 30 | 31 | #layer6 32 | x = MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_first' )(x) 33 | 34 | #layer7 35 | x = Conv2D(256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 36 | x = ( Activation('relu'))(x) 37 | x = ( BatchNormalization())(x) 38 | 39 | #layer8 40 | x = Conv2D(256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 41 | x = ( Activation('relu'))(x) 42 | x = ( BatchNormalization())(x) 43 | 44 | #layer9 45 | x = Conv2D(256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 46 | x = ( Activation('relu'))(x) 47 | x = ( BatchNormalization())(x) 48 | 49 | #layer10 50 | x = MaxPooling2D((2, 2), strides=(2, 2), data_format='channels_first' )(x) 51 | 52 | #layer11 53 | x = ( Conv2D(512, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 54 | x = ( Activation('relu'))(x) 55 | x = ( BatchNormalization())(x) 56 | 57 | #layer12 58 | x = ( Conv2D(512, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 59 | x = ( Activation('relu'))(x) 60 | x = ( BatchNormalization())(x) 61 | 62 | #layer13 63 | x = ( Conv2D(512, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 64 | x = ( Activation('relu'))(x) 65 | x = ( BatchNormalization())(x) 66 | 67 | #layer14 68 | x = ( UpSampling2D( (2,2), data_format='channels_first'))(x) 69 | 70 | #layer15 71 | x = ( Conv2D( 256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 72 | x = ( Activation('relu'))(x) 73 | x = ( BatchNormalization())(x) 74 | 75 | #layer16 76 | x = ( Conv2D( 256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 77 | x = ( Activation('relu'))(x) 78 | x = ( BatchNormalization())(x) 79 | 80 | #layer17 81 | x = ( Conv2D( 256, (3, 3), kernel_initializer='random_uniform', padding='same', data_format='channels_first'))(x) 82 | x = ( Activation('relu'))(x) 83 | x = ( BatchNormalization())(x) 84 | 85 | #layer18 86 | x = ( UpSampling2D( (2,2), data_format='channels_first'))(x) 87 | 88 | #layer19 89 | x = ( Conv2D( 128 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 90 | x = ( Activation('relu'))(x) 91 | x = ( BatchNormalization())(x) 92 | 93 | #layer20 94 | x = ( Conv2D( 128 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 95 | x = ( Activation('relu'))(x) 96 | x = ( BatchNormalization())(x) 97 | 98 | #layer21 99 | x = ( UpSampling2D( (2,2), data_format='channels_first'))(x) 100 | 101 | #layer22 102 | x = ( Conv2D( 64 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 103 | x = ( Activation('relu'))(x) 104 | x = ( BatchNormalization())(x) 105 | 106 | #layer23 107 | x = ( Conv2D( 64 , (3, 3), kernel_initializer='random_uniform', padding='same' , data_format='channels_first' ))(x) 108 | x = ( Activation('relu'))(x) 109 | x = ( BatchNormalization())(x) 110 | 111 | #layer24 112 | x = Conv2D( n_classes , (3, 3) , kernel_initializer='random_uniform', padding='same', data_format='channels_first' )(x) 113 | x = ( Activation('relu'))(x) 114 | x = ( BatchNormalization())(x) 115 | 116 | o_shape = Model(imgs_input , x ).output_shape 117 | print("layer24 output shape:", o_shape[1],o_shape[2],o_shape[3]) 118 | #layer24 output shape: 256, 360, 640 119 | 120 | OutputHeight = o_shape[2] 121 | OutputWidth = o_shape[3] 122 | 123 | #reshape the size to (256, 360*640) 124 | x = (Reshape(( -1 , OutputHeight*OutputWidth )))(x) 125 | 126 | #change dimension order to (360*640, 256) 127 | x = (Permute((2, 1)))(x) 128 | 129 | #layer25 130 | gaussian_output = (Activation('softmax'))(x) 131 | 132 | model = Model( imgs_input , gaussian_output) 133 | model.outputWidth = OutputWidth 134 | model.outputHeight = OutputHeight 135 | 136 | #show model's details 137 | model.summary() 138 | 139 | return model 140 | 141 | 142 | 143 | 144 | -------------------------------------------------------------------------------- /Code_Tennis/Models/__pycache__/TrackNet.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Tennis/Models/__pycache__/TrackNet.cpython-37.pyc -------------------------------------------------------------------------------- /Code_Tennis/Models/__pycache__/__init__.cpython-35.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Tennis/Models/__pycache__/__init__.cpython-35.pyc -------------------------------------------------------------------------------- /Code_Tennis/Models/__pycache__/__init__.cpython-37.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Tennis/Models/__pycache__/__init__.cpython-37.pyc -------------------------------------------------------------------------------- /Code_Tennis/predict.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import Models , LoadBatches 3 | import cv2 4 | import numpy as np 5 | import glob 6 | import os 7 | 8 | #parse parameters 9 | parser = argparse.ArgumentParser() 10 | parser.add_argument("--save_weights_path", type = str ) 11 | parser.add_argument("--test_images_path", type = str , default = "") 12 | parser.add_argument("--output_path", type = str , default = "") 13 | parser.add_argument("--input_height", type=int , default = 360 ) 14 | parser.add_argument("--input_width", type=int , default = 640 ) 15 | parser.add_argument("--output_height", type=int , default = 224 ) 16 | parser.add_argument("--output_width", type=int , default = 224 ) 17 | parser.add_argument("--n_classes", type=int ) 18 | 19 | args = parser.parse_args() 20 | n_classes = args.n_classes 21 | images_path = args.test_images_path 22 | output_path = args.output_path 23 | input_width = args.input_width 24 | input_height = args.input_height 25 | output_width = args.output_width 26 | output_height = args.output_height 27 | 28 | #load TrackNet model 29 | modelTN = Models.TrackNet.TrackNet 30 | m = modelTN( n_classes , input_height=input_height, input_width=input_width ) 31 | m.compile(loss='categorical_crossentropy', optimizer= 'adadelta', metrics=['accuracy']) 32 | m.load_weights( args.save_weights_path ) 33 | 34 | 35 | #get TrackNet output height and width 36 | model_output_height = m.outputHeight 37 | model_output_width = m.outputWidth 38 | 39 | #create grayscale RGB (0,0,0)~(255,255,255) 40 | colors = [ ( i, i, i ) for i in range(0, n_classes) ] 41 | 42 | #predict each clips from 1 to 82 43 | for clip in range(1,82): 44 | 45 | #get all JPG images in the path 46 | images = glob.glob( images_path + str(clip) + "/*.jpg" ) 47 | images.sort() 48 | 49 | 50 | #create folder for saving output image 51 | if not os.path.exists(output_path + str(clip) + "/"): 52 | os.makedirs(output_path + str(clip) + "/") 53 | 54 | #predict each images 55 | #since TrackNet cant predict first and second images, so we start from third image 56 | for i in range(2,len(images)): 57 | 58 | output_name = images[i].replace( images_path, output_path) 59 | 60 | #load input data 61 | X = LoadBatches.getInputArr( images[i], images[i-1], images[i-2], input_width, input_height ) 62 | #prdict heatmap 63 | pr = m.predict( np.array([X]) )[0] 64 | 65 | #since TrackNet output is ( net_output_height*model_output_width , n_classes ) 66 | #so we need to reshape image as ( net_output_height, model_output_width , n_classes(depth) ) 67 | pr = pr.reshape(( model_output_height, model_output_width , n_classes ) ).argmax( axis=2 ) 68 | 69 | #declare variable for output image 70 | output_img = np.zeros( ( model_output_height, model_output_width , 3 ) ) 71 | 72 | #set RGB value for each pixel predciton 73 | for c in range(n_classes): 74 | output_img[:,:,0] += ( (pr[:,: ] == c )*( colors[c][0] )).astype('uint8') 75 | output_img[:,:,1] += ((pr[:,: ] == c )*( colors[c][1] )).astype('uint8') 76 | output_img[:,:,2] += ((pr[:,: ] == c )*( colors[c][2] )).astype('uint8') 77 | 78 | #reshape the image size as original input image 79 | output_img = cv2.resize(output_img , (output_width , output_height )) 80 | 81 | #output heatmap image 82 | cv2.imwrite( output_name , output_img ) 83 | -------------------------------------------------------------------------------- /Code_Tennis/predict_video.py: -------------------------------------------------------------------------------- 1 | # python predict_video.py --save_weights_path=weights/model_tennis.h5 --input_video_path="tennis.mp4" --output_video_path="tracked.mp4" --n_classes=256 2 | 3 | import argparse 4 | # import Models 5 | from Models.TrackNet import TrackNet 6 | import queue 7 | import cv2 8 | import numpy as np 9 | from PIL import Image, ImageDraw 10 | 11 | #parse parameters 12 | parser = argparse.ArgumentParser() 13 | parser.add_argument("--input_video_path", type=str) 14 | parser.add_argument("--output_video_path", type=str, default = "") 15 | parser.add_argument("--save_weights_path", type = str ) 16 | parser.add_argument("--n_classes", type=int ) 17 | 18 | args = parser.parse_args() 19 | input_video_path = args.input_video_path 20 | output_video_path = args.output_video_path 21 | save_weights_path = args.save_weights_path 22 | n_classes = args.n_classes 23 | 24 | if output_video_path == "": 25 | #output video in same path 26 | output_video_path = input_video_path.split('.')[0] + "_TrackNet.mp4" 27 | 28 | #get video fps&video size 29 | video = cv2.VideoCapture(input_video_path) 30 | fps = int(video.get(cv2.CAP_PROP_FPS)) 31 | output_width = int(video.get(cv2.CAP_PROP_FRAME_WIDTH)) 32 | output_height = int(video.get(cv2.CAP_PROP_FRAME_HEIGHT)) 33 | 34 | #start from first frame 35 | currentFrame = 0 36 | 37 | #width and height in TrackNet 38 | width , height = 640, 360 39 | img, img1, img2 = None, None, None 40 | 41 | #load TrackNet model 42 | modelFN = TrackNet 43 | m = modelFN( n_classes , input_height=height, input_width=width ) 44 | m.compile(loss='categorical_crossentropy', optimizer= 'adadelta' , metrics=['accuracy']) 45 | m.load_weights( save_weights_path ) 46 | 47 | # In order to draw the trajectory of tennis, we need to save the coordinate of previous 7 frames 48 | q = queue.deque() 49 | for i in range(0,8): 50 | q.appendleft(None) 51 | 52 | #save prediction images as vidoe 53 | #Tutorial: https://stackoverflow.com/questions/33631489/error-during-saving-a-video-using-python-and-opencv 54 | fourcc = cv2.VideoWriter_fourcc(*'XVID') 55 | output_video = cv2.VideoWriter(output_video_path,fourcc, fps, (output_width,output_height)) 56 | 57 | 58 | #both first and second frames cant be predict, so we directly write the frames to output video 59 | #capture frame-by-frame 60 | video.set(1,currentFrame); 61 | ret, img1 = video.read() 62 | #write image to video 63 | output_video.write(img1) 64 | currentFrame +=1 65 | #resize it 66 | img1 = cv2.resize(img1, ( width , height )) 67 | #input must be float type 68 | img1 = img1.astype(np.float32) 69 | 70 | #capture frame-by-frame 71 | video.set(1,currentFrame); 72 | ret, img = video.read() 73 | #write image to video 74 | output_video.write(img) 75 | currentFrame +=1 76 | #resize it 77 | img = cv2.resize(img, ( width , height )) 78 | #input must be float type 79 | img = img.astype(np.float32) 80 | 81 | 82 | 83 | while(True): 84 | 85 | img2 = img1 86 | img1 = img 87 | 88 | #capture frame-by-frame 89 | video.set(1,currentFrame); 90 | ret, img = video.read() 91 | 92 | #if there dont have any frame in video, break 93 | if not ret: 94 | break 95 | 96 | #img is the frame that TrackNet will predict the position 97 | #since we need to change the size and type of img, copy it to output_img 98 | output_img = img 99 | 100 | #resize it 101 | img = cv2.resize(img, ( width , height )) 102 | #input must be float type 103 | img = img.astype(np.float32) 104 | 105 | 106 | #combine three imgs to (width , height, rgb*3) 107 | X = np.concatenate((img, img1, img2),axis=2) 108 | 109 | #since the odering of TrackNet is 'channels_first', so we need to change the axis 110 | X = np.rollaxis(X, 2, 0) 111 | #prdict heatmap 112 | pr = m.predict( np.array([X]) )[0] 113 | 114 | #since TrackNet output is ( net_output_height*model_output_width , n_classes ) 115 | #so we need to reshape image as ( net_output_height, model_output_width , n_classes(depth) ) 116 | #.argmax( axis=2 ) => select the largest probability as class 117 | pr = pr.reshape(( height , width , n_classes ) ).argmax( axis=2 ) 118 | 119 | #cv2 image must be numpy.uint8, convert numpy.int64 to numpy.uint8 120 | pr = pr.astype(np.uint8) 121 | 122 | #reshape the image size as original input image 123 | heatmap = cv2.resize(pr , (output_width, output_height )) 124 | 125 | #heatmap is converted into a binary image by threshold method. 126 | ret,heatmap = cv2.threshold(heatmap,127,255,cv2.THRESH_BINARY) 127 | 128 | #find the circle in image with 2<=radius<=7 129 | circles = cv2.HoughCircles(heatmap, cv2.HOUGH_GRADIENT,dp=1,minDist=1,param1=50,param2=2,minRadius=2,maxRadius=7) 130 | 131 | #In order to draw the circle in output_img, we need to used PIL library 132 | #Convert opencv image format to PIL image format 133 | PIL_image = cv2.cvtColor(output_img, cv2.COLOR_BGR2RGB) 134 | PIL_image = Image.fromarray(PIL_image) 135 | 136 | #check if there have any tennis be detected 137 | if circles is not None: 138 | #if only one tennis be detected 139 | if len(circles) == 1: 140 | 141 | x = int(circles[0][0][0]) 142 | y = int(circles[0][0][1]) 143 | print(currentFrame, x,y) 144 | 145 | #push x,y to queue 146 | q.appendleft([x,y]) 147 | #pop x,y from queue 148 | q.pop() 149 | else: 150 | #push None to queue 151 | q.appendleft(None) 152 | #pop x,y from queue 153 | q.pop() 154 | else: 155 | #push None to queue 156 | q.appendleft(None) 157 | #pop x,y from queue 158 | q.pop() 159 | 160 | #draw current frame prediction and previous 7 frames as yellow circle, total: 8 frames 161 | for i in range(0,8): 162 | if q[i] is not None: 163 | draw_x = q[i][0] 164 | draw_y = q[i][1] 165 | bbox = (draw_x - 2, draw_y - 2, draw_x + 2, draw_y + 2) 166 | draw = ImageDraw.Draw(PIL_image) 167 | draw.ellipse(bbox, outline ='yellow') 168 | del draw 169 | 170 | #Convert PIL image format back to opencv image format 171 | opencvImage = cv2.cvtColor(np.array(PIL_image), cv2.COLOR_RGB2BGR) 172 | #write image to output_video 173 | output_video.write(opencvImage) 174 | 175 | #next frame 176 | currentFrame += 1 177 | 178 | # everything is done, release the video 179 | video.release() 180 | output_video.release() 181 | print("finish") -------------------------------------------------------------------------------- /Code_Tennis/train.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | import Models , LoadBatches 3 | from keras import optimizers 4 | from keras.utils import plot_model 5 | 6 | #parse parameters 7 | parser = argparse.ArgumentParser() 8 | parser.add_argument("--save_weights_path", type = str ) 9 | parser.add_argument("--training_images_name", type = str ) 10 | parser.add_argument("--n_classes", type=int ) 11 | parser.add_argument("--input_height", type=int , default = 360 ) 12 | parser.add_argument("--input_width", type=int , default = 640 ) 13 | parser.add_argument("--epochs", type = int, default = 1000 ) 14 | parser.add_argument("--batch_size", type = int, default = 2 ) 15 | parser.add_argument("--load_weights", type = str , default = "-1") 16 | parser.add_argument("--step_per_epochs", type = int, default = 200 ) 17 | 18 | args = parser.parse_args() 19 | training_images_name = args.training_images_name 20 | train_batch_size = args.batch_size 21 | n_classes = args.n_classes 22 | input_height = args.input_height 23 | input_width = args.input_width 24 | save_weights_path = args.save_weights_path 25 | epochs = args.epochs 26 | load_weights = args.load_weights 27 | step_per_epochs = args.step_per_epochs 28 | optimizer_name = optimizers.Adadelta(lr=1.0) 29 | 30 | #load TrackNet model 31 | modelTN = Models.TrackNet.TrackNet 32 | m = modelTN( n_classes , input_height=input_height, input_width=input_width ) 33 | m.compile(loss='categorical_crossentropy', optimizer= optimizer_name, metrics=['accuracy']) 34 | 35 | #check if need to retrain the model weights 36 | if load_weights != "-1": 37 | m.load_weights("weights/model." + load_weights) 38 | 39 | #show TrackNet details, save it as TrackNet.png 40 | plot_model( m , show_shapes=True , to_file='TrackNet.png') 41 | 42 | #get TrackNet output height and width 43 | model_output_height = m.outputHeight 44 | model_output_width = m.outputWidth 45 | 46 | #creat input data and output data 47 | Generator = LoadBatches.InputOutputGenerator( training_images_name, train_batch_size, n_classes , input_height , input_width , model_output_height , model_output_width) 48 | 49 | 50 | #start to train the model, and save weights until finish 51 | ''' 52 | m.fit_generator( Generator, step_per_epochs, epochs ) 53 | m.save_weights( save_weights_path + ".0" ) 54 | 55 | ''' 56 | 57 | #start to train the model, and save weights per 50 epochs 58 | 59 | for ep in range(1, epochs+1 ): 60 | print "Epoch :", str(ep) + "/" + str(epochs) 61 | m.fit_generator(Generator, step_per_epochs) 62 | if ep % 50 == 0: 63 | m.save_weights(save_weights_path + ".0") 64 | -------------------------------------------------------------------------------- /Code_Tennis/weights/model_tennis.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Code_Tennis/weights/model_tennis.h5 -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool.fig: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool.fig -------------------------------------------------------------------------------- /LabelingTool/LabelingTool.m: -------------------------------------------------------------------------------- 1 | function varargout = LabelingTool(varargin) 2 | gui_Singleton = 1; 3 | gui_State = struct('gui_Name', mfilename, ... 4 | 'gui_Singleton', gui_Singleton, ... 5 | 'gui_OpeningFcn', @LabelingTool_OpeningFcn, ... 6 | 'gui_OutputFcn', @LabelingTool_OutputFcn, ... 7 | 'gui_LayoutFcn', [] , ... 8 | 'gui_Callback', []); 9 | if nargin && ischar(varargin{1}) 10 | gui_State.gui_Callback = str2func(varargin{1}); 11 | end 12 | 13 | if nargout 14 | [varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:}); 15 | else 16 | gui_mainfcn(gui_State, varargin{:}); 17 | end 18 | % End initialization code - DO NOT EDIT 19 | end 20 | 21 | % --- Outputs from this function are returned to the command line. 22 | function varargout = LabelingTool_OutputFcn(hObject, eventdata, handles) 23 | varargout{1} = handles.output; 24 | 25 | % initialization code 26 | end 27 | 28 | 29 | % --- Executes just before LabelingTool is made visible. 30 | function LabelingTool_OpeningFcn(hObject, eventdata, handles, varargin) 31 | handles.output = hObject; 32 | set(gcf,'pointer','fullcrosshair') 33 | guidata(hObject, handles); 34 | 35 | end 36 | 37 | 38 | % --- Executes on button press in pushbutton1. 39 | function pushbutton1_Callback(hObject, eventdata, handles) 40 | global idx imgs Visibility Status CoordinateX CoordinateY draw_circle; 41 | 42 | dName_img = uigetdir(); 43 | if dName_img ~= 0 44 | set(handles.text2, 'String', dName_img); % set path on ui 45 | 46 | idx = 1;% current frame index 47 | CoordinateX(1:10000)= -1; % Array to save each frame Coordinate X 48 | CoordinateY(1:10000)= -1; % Array to save each frame Coordinate Y 49 | Status(1:10000)= 0; % Array to save each frame status 50 | Visibility(1:10000)= 1; % Array to save each frame visibility 51 | draw_circle = false; % Flag to check if the point be show on axes 52 | 53 | % save all files path string array 54 | imgs = getAllFiles(dName_img); 55 | guidata(hObject, handles); 56 | 57 | 58 | axes(handles.axes1); % set current axes to axes1 59 | %show image on axes 60 | imageHandle = imshow(imgs{idx}); 61 | 62 | % assign the ButtonDownFcn to the image handle 63 | set(imageHandle,'ButtonDownFcn',@ImageClickCallback); 64 | 65 | % show path on ui 66 | set(handles.text3, 'String', imgs{idx}); 67 | 68 | %check if Label.csv in this path, if yes, load and show the data 69 | if exist(strcat(dName_img,'/Label.csv'), 'file') == 2 % value 2 means file exist 70 | fid = fopen(strcat(dName_img,'/Label.csv')); 71 | out = textscan(fid,'%s%d%f%f%d','Headerlines',1,'delimiter',','); 72 | fclose(fid); 73 | 74 | for i = 1:size(imgs) 75 | Visibility(i) = out{2}(i); 76 | CoordinateX(i) = out{3}(i); 77 | CoordinateY(i) = out{4}(i); 78 | Status(i) = out{5}(i); 79 | end 80 | end 81 | end 82 | 83 | end 84 | 85 | 86 | % Get mouse click coordinates function, tutorial: https://stackoverflow.com/questions/14684577/matlab-how-to-get-mouse-click-coordinates/14685313 87 | function ImageClickCallback ( objectHandle , eventData) 88 | 89 | global idx CoordinateX CoordinateY draw_circle Circle 90 | 91 | axesHandle = get(objectHandle,'Parent'); 92 | 93 | % get mouse click coordinates x, y 94 | coordinates = get(axesHandle,'CurrentPoint'); 95 | x = coordinates(1,1); 96 | y = coordinates(1,2); 97 | 98 | % save x, y to global variable Coordinate 99 | CoordinateX(idx) = round(x); 100 | CoordinateY(idx) = round(y); 101 | 102 | % since only a circle can be plot in axes, so we need to check if we had draw the circle in axes, if yes, we need to delete it 103 | radius = 1; 104 | if draw_circle == true 105 | delete(Circle); 106 | end 107 | 108 | %draw the circle on axes 109 | Circle = viscircles([x y],radius,'Color','r'); 110 | draw_circle = true; 111 | 112 | end 113 | 114 | 115 | 116 | % get all *.jpg file function, refer to https://cn.mathworks.com/matlabcentral/fileexchange/48238-nonstationary-extreme-value-analysis--neva--toolbox?focused=5028815&tab=function 117 | function fileList = getAllFiles(dirName) 118 | dirData = dir(strcat(dirName,'/*.jpg')); 119 | dirIndex = [dirData.isdir]; 120 | fileList = {dirData(~dirIndex).name}'; 121 | 122 | if ~isempty(fileList) 123 | fileList = cellfun(@(x) fullfile(dirName,x),fileList,'UniformOutput',false); 124 | end 125 | 126 | subDirs = {dirData(dirIndex).name}; 127 | validIndex = ~ismember(subDirs,{'.','..'}); 128 | 129 | for iDir = find(validIndex) 130 | nextDir = fullfile(dirName,subDirs{iDir}); 131 | fileList = [fileList; getAllFiles(nextDir)]; 132 | end 133 | 134 | end 135 | 136 | 137 | 138 | % --- Executes on key press with focus on figure1 139 | function figure1_WindowKeyPressFcn(hObject, eventdata, handles) 140 | 141 | global idx imgs draw_circle CoordinateX CoordinateY Circle Visibility Status 142 | 143 | switch eventdata.Key 144 | case 'leftarrow' % show previous frame 145 | if idx > 1 % idx range should be 1~lenght(imgs) 146 | idx = idx - 1; 147 | guidata(hObject, handles); 148 | % set current axes to axes1 149 | axes(handles.axes1); 150 | %show image on axes 151 | imageHandle = imshow(imgs{idx}); 152 | % assign the ButtonDownFcn to the image handle 153 | set(imageHandle,'ButtonDownFcn',@ImageClickCallback); 154 | 155 | % default all of radio buttons is enable 156 | set(handles.frying, 'Enable', 'on'); 157 | set(handles.hit, 'Enable', 'on'); 158 | set(handles.bouncing, 'Enable', 'on'); 159 | 160 | 161 | if Visibility(idx) == 0 % disable all of radio buttons 162 | set(handles.no_ball, 'Value', 1); 163 | set(handles.frying, 'Enable', 'off'); 164 | set(handles.hit, 'Enable', 'off'); 165 | set(handles.bouncing, 'Enable', 'off'); 166 | elseif Visibility(idx) == 1 % set easy_identification radio button to 1 167 | set(handles.easy_identification, 'Value', 1); 168 | elseif Visibility(idx) == 2 % set hard_identification radio button to 1 169 | set(handles.hard_identification, 'Value', 1); 170 | elseif Visibility(idx) == 3 % set occluded_ball radio button to 1 171 | set(handles.occluded_ball, 'Value', 1); 172 | end 173 | 174 | if Status(idx) == 0 % set frying radio button to 1 175 | set(handles.frying, 'Value', 1); 176 | elseif Status(idx) == 1 % set hit radio button to 1 177 | set(handles.hit, 'Value', 1); 178 | elseif Status(idx) == 2 % set bouncing radio button to 1 179 | set(handles.bouncing, 'Value', 1); 180 | end 181 | 182 | set(handles.text3, 'String', imgs{idx}); % show path on text3 183 | 184 | if CoordinateX(idx)==-1 && CoordinateY(idx)==-1 185 | set(handles.text3, 'ForegroundColor', 'r'); % for reminding, if the frame do not be labeled, the path will be red. 186 | draw_circle = false; 187 | else 188 | set(handles.text3, 'ForegroundColor', 'b'); % for reminding, if the frame do not be labeled, the path will be red 189 | Circle = viscircles([CoordinateX(idx) CoordinateY(idx) ],1,'Color','r'); 190 | draw_circle = true; 191 | end 192 | end 193 | 194 | case 'rightarrow' % show next frame 195 | [m,n] = size(imgs); % get images size 196 | if idx < m % idx range should be 1~lenght(imgs) 197 | 198 | idx = idx + 1; 199 | guidata(hObject, handles); 200 | % set current axes to axes1 201 | axes(handles.axes1); 202 | %show image on axes 203 | imageHandle = imshow(imgs{idx}); 204 | % assign the ButtonDownFcn to the image handle 205 | set(imageHandle,'ButtonDownFcn',@ImageClickCallback); 206 | 207 | % default all of radio buttons is enable 208 | set(handles.frying, 'Enable', 'on'); 209 | set(handles.hit, 'Enable', 'on'); 210 | set(handles.bouncing, 'Enable', 'on'); 211 | 212 | if Visibility(idx) == 0 % disable all of radio buttons 213 | set(handles.no_ball, 'Value', 1); 214 | set(handles.frying, 'Enable', 'off'); 215 | set(handles.hit, 'Enable', 'off'); 216 | set(handles.bouncing, 'Enable', 'off'); 217 | elseif Visibility(idx) == 1 % set easy_identification radio button to 1 218 | set(handles.easy_identification, 'Value', 1); 219 | elseif Visibility(idx) == 2 % set hard_identification radio button to 1 220 | set(handles.hard_identification, 'Value', 1); 221 | elseif Visibility(idx) == 3 % set occluded_ball radio button to 1 222 | set(handles.occluded_ball, 'Value', 1); 223 | end 224 | 225 | if Status(idx) == 0 % set frying radio button to 1 226 | set(handles.frying, 'Value', 1); 227 | elseif Status(idx) == 1 % set hit radio button to 1 228 | set(handles.hit, 'Value', 1); 229 | elseif Status(idx) == 2 % set bouncing radio button to 1 230 | set(handles.bouncing, 'Value', 1); 231 | end 232 | 233 | 234 | set(handles.text3, 'String', imgs{idx}); % show path on text3 235 | if CoordinateX(idx)==-1 && CoordinateY(idx)==-1 236 | set(handles.text3, 'ForegroundColor', 'r'); % for reminding, if the frame do not be labeled, the path will be as red. 237 | draw_circle = false; 238 | else 239 | set(handles.text3, 'ForegroundColor', 'b'); % for reminding, if the frame do not be labeled, the path will be as blue. 240 | Circle = viscircles([CoordinateX(idx) CoordinateY(idx) ],1,'Color','r'); 241 | draw_circle = true; 242 | end 243 | end 244 | end 245 | 246 | end 247 | 248 | 249 | % --- Executes on button press in pushbutton2. 250 | function pushbutton2_Callback(hObject, eventdata, handles) 251 | 252 | global imgs CoordinateX CoordinateY Visibility Status 253 | 254 | % for reminding, check if there have Coordinates dont be labeled 255 | alert = false; 256 | for i = 1:size(imgs) 257 | if Visibility(i) ~= 0 258 | if CoordinateX(i)==-1 || CoordinateY(i) ==-1 259 | alert = true; 260 | Path = strsplit(char(imgs(i)),'/'); 261 | fileName = Path{length(Path)}; 262 | warndlg(strcat(fileName, 'Visibility is not No BALL, but Coordinates dont be labeled') ,'Alert'); 263 | break; 264 | end 265 | end 266 | end 267 | 268 | % if all of images be labeled correctly, output Label.csv 269 | if alert == false 270 | disp(strcat(get(handles.text2, 'String'))) 271 | fid = fopen(strcat(get(handles.text2, 'String'),'\Label.csv'),'w'); 272 | fprintf(fid, 'file name,visibility,x-coordinate,y-coordinate,status\n'); % Label.csv header 273 | for i = 1:size(imgs) 274 | Path = strsplit(char(imgs(i)),'\'); 275 | fileName = Path{length(Path)}; 276 | if Visibility(i)== 0 277 | fprintf(fid, '%s,0,,,\n', fileName); 278 | else 279 | fprintf(fid, '%s,%d,%d,%d,%d\n', fileName, Visibility(i), CoordinateX(i), CoordinateY(i), Status(i)); 280 | 281 | end 282 | end 283 | fclose(fid); 284 | end 285 | 286 | 287 | end 288 | 289 | 290 | % --- Executes when selected object is changed in VisibilityGroup. 291 | function VisibilityGroup_SelectionChangedFcn(hObject, eventdata, handles) 292 | global idx Visibility 293 | 294 | % default all of radiobutton as enable 295 | set(handles.frying, 'Enable', 'on'); 296 | set(handles.hit, 'Enable', 'on'); 297 | set(handles.bouncing, 'Enable', 'on'); 298 | 299 | switch(get(eventdata.NewValue,'Tag')); 300 | case 'no_ball' % turn all of radiobutton as disable 301 | Visibility(idx) = 0; 302 | set(handles.frying, 'Enable', 'off'); 303 | set(handles.hit, 'Enable', 'off'); 304 | set(handles.bouncing, 'Enable', 'off'); 305 | case 'easy_identification' 306 | Visibility(idx) = 1; 307 | case 'hard_identification' 308 | Visibility(idx) = 2; 309 | case 'occluded_ball' 310 | Visibility(idx) = 3; 311 | end 312 | 313 | end 314 | 315 | 316 | % --- Executes when selected object is changed in StatusGroup. 317 | function StatusGroup_SelectionChangedFcn(hObject, eventdata, handles) 318 | global idx Status 319 | switch(get(eventdata.NewValue,'Tag')); 320 | case 'frying' 321 | Status(idx) = 0; 322 | case 'hit' 323 | Status(idx) = 1; 324 | case 'bouncing' 325 | Status(idx) = 2; 326 | end 327 | end 328 | 329 | 330 | 331 | % --- Executes on key press with focus on figure1 or any of its controls. 332 | function figure1_KeyPressFcn(hObject, eventdata, handles) 333 | end 334 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool.m~: -------------------------------------------------------------------------------- 1 | function varargout = LabelingTool(varargin) 2 | gui_Singleton = 1; 3 | gui_State = struct('gui_Name', mfilename, ... 4 | 'gui_Singleton', gui_Singleton, ... 5 | 'gui_OpeningFcn', @LabelingTool_OpeningFcn, ... 6 | 'gui_OutputFcn', @LabelingTool_OutputFcn, ... 7 | 'gui_LayoutFcn', [] , ... 8 | 'gui_Callback', []); 9 | if nargin && ischar(varargin{1}) 10 | gui_State.gui_Callback = str2func(varargin{1}); 11 | end 12 | 13 | if nargout 14 | [varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:}); 15 | else 16 | gui_mainfcn(gui_State, varargin{:}); 17 | end 18 | % End initialization code - DO NOT EDIT 19 | end 20 | 21 | % --- Outputs from this function are returned to the command line. 22 | function varargout = LabelingTool_OutputFcn(hObject, eventdata, handles) 23 | varargout{1} = handles.output; 24 | 25 | % initialization code 26 | end 27 | 28 | 29 | % --- Executes just before LabelingTool is made visible. 30 | function LabelingTool_OpeningFcn(hObject, eventdata, handles, varargin) 31 | handles.output = hObject; 32 | set(gcf,'pointer','fullcrosshair') 33 | guidata(hObject, handles); 34 | 35 | end 36 | 37 | 38 | % --- Executes on button press in pushbutton1. 39 | function pushbutton1_Callback(hObject, eventdata, handles) 40 | global idx imgs Visibility Status CoordinateX CoordinateY draw_circle; 41 | 42 | dName_img = uigetdir('/media/andersen/D/Clip1'); 43 | if dName_img ~= 0 44 | set(handles.text2, 'String', dName_img); % set path on ui 45 | 46 | idx = 1;% current frame index 47 | CoordinateX(1:10000)= -1; % Array to save each frame Coordinate X 48 | CoordinateY(1:10000)= -1; % Array to save each frame Coordinate Y 49 | Status(1:10000)= 0; % Array to save each frame status 50 | Visibility(1:10000)= 1; % Array to save each frame visibility 51 | draw_circle = false; % Flag to check if the point be show on axes 52 | 53 | % save all files path string array 54 | imgs = getAllFiles(dName_img); 55 | guidata(hObject, handles); 56 | 57 | 58 | axes(handles.axes1); % set current axes to axes1 59 | %show image on axes 60 | imageHandle = imshow(imgs{idx}); 61 | 62 | % assign the ButtonDownFcn to the image handle 63 | set(imageHandle,'ButtonDownFcn',@ImageClickCallback); 64 | 65 | % show path on ui 66 | set(handles.text3, 'String', imgs{idx}); 67 | 68 | %check if Label.csv in this path, if yes, load and show the data 69 | if exist(strcat(dName_img,'/Label.csv'), 'file') == 2 % value 2 means file exist 70 | fid = fopen(strcat(dName_img,'/Label.csv')); 71 | out = textscan(fid,'%s%d%f%f%d','Headerlines',1,'delimiter',','); 72 | fclose(fid); 73 | 74 | for i = 1:size(imgs) 75 | Visibility(i) = out{2}(i); 76 | CoordinateX(i) = out{3}(i); 77 | CoordinateY(i) = out{4}(i); 78 | Status(i) = out{5}(i); 79 | end 80 | end 81 | end 82 | 83 | end 84 | 85 | 86 | % Get mouse click coordinates function, tutorial: https://stackoverflow.com/questions/14684577/matlab-how-to-get-mouse-click-coordinates/14685313 87 | function ImageClickCallback ( objectHandle , eventData) 88 | 89 | global idx CoordinateX CoordinateY draw_circle Circle 90 | 91 | axesHandle = get(objectHandle,'Parent'); 92 | 93 | % get mouse click coordinates x, y 94 | coordinates = get(axesHandle,'CurrentPoint'); 95 | x = coordinates(1,1); 96 | y = coordinates(1,2); 97 | 98 | % save x, y to global variable Coordinate 99 | CoordinateX(idx) = round(x); 100 | CoordinateY(idx) = round(y); 101 | 102 | % since only a circle can be plot in axes, so we need to check if we had draw the circle in axes, if yes, we need to delete it 103 | radius = 1; 104 | if draw_circle == true 105 | delete(Circle); 106 | end 107 | 108 | %draw the circle on axes 109 | Circle = viscircles([x y],radius,'Color','r'); 110 | draw_circle = true; 111 | 112 | end 113 | 114 | 115 | 116 | % get all *.jpg file function, refer to https://cn.mathworks.com/matlabcentral/fileexchange/48238-nonstationary-extreme-value-analysis--neva--toolbox?focused=5028815&tab=function 117 | function fileList = getAllFiles(dirName) 118 | dirData = dir(strcat(dirName,'/*.jpg')); 119 | dirIndex = [dirData.isdir]; 120 | fileList = {dirData(~dirIndex).name}'; 121 | 122 | if ~isempty(fileList) 123 | fileList = cellfun(@(x) fullfile(dirName,x),fileList,'UniformOutput',false); 124 | end 125 | 126 | subDirs = {dirData(dirIndex).name}; 127 | validIndex = ~ismember(subDirs,{'.','..'}); 128 | 129 | for iDir = find(validIndex) 130 | nextDir = fullfile(dirName,subDirs{iDir}); 131 | fileList = [fileList; getAllFiles(nextDir)]; 132 | end 133 | 134 | end 135 | 136 | 137 | 138 | % --- Executes on key press with focus on figure1 139 | function figure1_WindowKeyPressFcn(hObject, eventdata, handles) 140 | 141 | global idx imgs draw_circle CoordinateX CoordinateY Circle Visibility Status 142 | 143 | switch eventdata.Key 144 | case 'leftarrow' % show previous frame 145 | if idx > 1 % idx range should be 1~lenght(imgs) 146 | idx = idx - 1; 147 | guidata(hObject, handles); 148 | % set current axes to axes1 149 | axes(handles.axes1); 150 | %show image on axes 151 | imageHandle = imshow(imgs{idx}); 152 | % assign the ButtonDownFcn to the image handle 153 | set(imageHandle,'ButtonDownFcn',@ImageClickCallback); 154 | 155 | % default all of radio buttons is enable 156 | set(handles.frying, 'Enable', 'on'); 157 | set(handles.hit, 'Enable', 'on'); 158 | set(handles.bouncing, 'Enable', 'on'); 159 | 160 | 161 | if Visibility(idx) == 0 % disable all of radio buttons 162 | set(handles.no_ball, 'Value', 1); 163 | set(handles.frying, 'Enable', 'off'); 164 | set(handles.hit, 'Enable', 'off'); 165 | set(handles.bouncing, 'Enable', 'off'); 166 | elseif Visibility(idx) == 1 % set easy_identification radio button to 1 167 | set(handles.easy_identification, 'Value', 1); 168 | elseif Visibility(idx) == 2 % set hard_identification radio button to 1 169 | set(handles.hard_identification, 'Value', 1); 170 | elseif Visibility(idx) == 3 % set occluded_ball radio button to 1 171 | set(handles.occluded_ball, 'Value', 1); 172 | end 173 | 174 | if Status(idx) == 0 % set frying radio button to 1 175 | set(handles.frying, 'Value', 1); 176 | elseif Status(idx) == 1 % set hit radio button to 1 177 | set(handles.hit, 'Value', 1); 178 | elseif Status(idx) == 2 % set bouncing radio button to 1 179 | set(handles.bouncing, 'Value', 1); 180 | end 181 | 182 | set(handles.text3, 'String', imgs{idx}); % show path on text3 183 | 184 | if CoordinateX(idx)==-1 && CoordinateY(idx)==-1 185 | set(handles.text3, 'ForegroundColor', 'r'); % for reminding, if the frame do not be labeled, the path will be red. 186 | draw_circle = false; 187 | else 188 | set(handles.text3, 'ForegroundColor', 'b'); % for reminding, if the frame do not be labeled, the path will be red 189 | Circle = viscircles([CoordinateX(idx) CoordinateY(idx) ],1,'Color','r'); 190 | draw_circle = true; 191 | end 192 | end 193 | 194 | case 'rightarrow' % show next frame 195 | [m,n] = size(imgs); % get images size 196 | if idx < m % idx range should be 1~lenght(imgs) 197 | 198 | idx = idx + 1; 199 | guidata(hObject, handles); 200 | % set current axes to axes1 201 | axes(handles.axes1); 202 | %show image on axes 203 | imageHandle = imshow(imgs{idx}); 204 | % assign the ButtonDownFcn to the image handle 205 | set(imageHandle,'ButtonDownFcn',@ImageClickCallback); 206 | 207 | % default all of radio buttons is enable 208 | set(handles.frying, 'Enable', 'on'); 209 | set(handles.hit, 'Enable', 'on'); 210 | set(handles.bouncing, 'Enable', 'on'); 211 | 212 | if Visibility(idx) == 0 % disable all of radio buttons 213 | set(handles.no_ball, 'Value', 1); 214 | set(handles.frying, 'Enable', 'off'); 215 | set(handles.hit, 'Enable', 'off'); 216 | set(handles.bouncing, 'Enable', 'off'); 217 | elseif Visibility(idx) == 1 % set easy_identification radio button to 1 218 | set(handles.easy_identification, 'Value', 1); 219 | elseif Visibility(idx) == 2 % set hard_identification radio button to 1 220 | set(handles.hard_identification, 'Value', 1); 221 | elseif Visibility(idx) == 3 % set occluded_ball radio button to 1 222 | set(handles.occluded_ball, 'Value', 1); 223 | end 224 | 225 | if Status(idx) == 0 % set frying radio button to 1 226 | set(handles.frying, 'Value', 1); 227 | elseif Status(idx) == 1 % set hit radio button to 1 228 | set(handles.hit, 'Value', 1); 229 | elseif Status(idx) == 2 % set bouncing radio button to 1 230 | set(handles.bouncing, 'Value', 1); 231 | end 232 | 233 | 234 | set(handles.text3, 'String', imgs{idx}); % show path on text3 235 | if CoordinateX(idx)==-1 && CoordinateY(idx)==-1 236 | set(handles.text3, 'ForegroundColor', 'r'); % for reminding, if the frame do not be labeled, the path will be as red. 237 | draw_circle = false; 238 | else 239 | set(handles.text3, 'ForegroundColor', 'b'); % for reminding, if the frame do not be labeled, the path will be as blue. 240 | Circle = viscircles([CoordinateX(idx) CoordinateY(idx) ],1,'Color','r'); 241 | draw_circle = true; 242 | end 243 | end 244 | end 245 | 246 | end 247 | 248 | 249 | % --- Executes on button press in pushbutton2. 250 | function pushbutton2_Callback(hObject, eventdata, handles) 251 | 252 | global imgs CoordinateX CoordinateY Visibility Status 253 | 254 | % for reminding, check if there have Coordinates dont be labeled 255 | alert = false; 256 | for i = 1:size(imgs) 257 | if Visibility(i) ~= 0 258 | if CoordinateX(i)==-1 || CoordinateY(i) ==-1 259 | alert = true; 260 | Path = strsplit(char(imgs(i)),'/'); 261 | fileName = Path{length(Path)}; 262 | warndlg(strcat(fileName, 'Visibility is not No BALL, but Coordinates dont be labeled') ,'Alert'); 263 | break; 264 | end 265 | end 266 | end 267 | 268 | % if all of images be labeled correctly, output Label.csv 269 | if alert == false 270 | fid = fopen(strcat(get(handles.text2, 'String'),'/Label.csv'),'w'); 271 | fprintf(fid, 'file name,visibility,x-coordinate,y-coordinate,status\n'); % Label.csv header 272 | for i = 1:size(imgs) 273 | Path = strsplit(char(imgs(i)),'/'); 274 | fileName = Path{length(Path)}; 275 | if Visibility(i)== 0 276 | fprintf(fid, '%s,0,,,\n', fileName); 277 | else 278 | fprintf(fid, '%s,%d,%d,%d,%d\n', fileName, Visibility(i), CoordinateX(i), CoordinateY(i), Status(i)); 279 | 280 | end 281 | end 282 | fclose(fid); 283 | end 284 | 285 | 286 | end 287 | 288 | 289 | % --- Executes when selected object is changed in VisibilityGroup. 290 | function VisibilityGroup_SelectionChangedFcn(hObject, eventdata, handles) 291 | global idx Visibility 292 | 293 | % default all of radiobutton as enable 294 | set(handles.frying, 'Enable', 'on'); 295 | set(handles.hit, 'Enable', 'on'); 296 | set(handles.bouncing, 'Enable', 'on'); 297 | 298 | switch(get(eventdata.NewValue,'Tag')); 299 | case 'no_ball' 300 | Visibility(idx) = 0; 301 | set(handles.frying, 'Enable', 'off'); 302 | set(handles.hit, 'Enable', 'off'); 303 | set(handles.bouncing, 'Enable', 'off'); 304 | case 'easy_identification' 305 | Visibility(idx) = 1; 306 | case 'hard_identification' 307 | Visibility(idx) = 2; 308 | case 'occluded_ball' 309 | Visibility(idx) = 3; 310 | end 311 | 312 | end 313 | 314 | 315 | % --- Executes when selected object is changed in StatusGroup. 316 | function StatusGroup_SelectionChangedFcn(hObject, eventdata, handles) 317 | global idx Status 318 | 319 | switch(get(eventdata.NewValue,'Tag')); 320 | case 'frying' 321 | Status(idx) = 0; 322 | case 'hit' 323 | Status(idx) = 1; 324 | case 'bouncing' 325 | Status(idx) = 2; 326 | end 327 | end 328 | 329 | 330 | 331 | % --- Executes on key press with focus on figure1 or any of its controls. 332 | function figure1_KeyPressFcn(hObject, eventdata, handles) 333 | end -------------------------------------------------------------------------------- /LabelingTool/LabelingTool.prj: -------------------------------------------------------------------------------- 1 | 2 | 3 | LabelingTool 4 | 5 | 6 | 1.0 7 | NOL 8 | 9 | NCTU 10 | 11 | 12 | 13 | 14 | \NCTU\LabelingTool\ 15 | option.installpath.programfiles 16 | 17 | 18 | 19 | ${PROJECT_ROOT}\LabelingTool\for_testing 20 | ${PROJECT_ROOT}\LabelingTool\for_redistribution_files_only 21 | ${PROJECT_ROOT}\LabelingTool\for_redistribution 22 | ${PROJECT_ROOT}\LabelingTool 23 | false 24 | 25 | subtarget.standalone 26 | 27 | true 28 | false 29 | false 30 | MyAppInstaller_web 31 | MyAppInstaller_mcr 32 | MyAppInstaller_app 33 | true 34 | false 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | ${PROJECT_ROOT}\LabelingTool.m 69 | 70 | 71 | ${PROJECT_ROOT}\LabelingTool.fig 72 | 73 | 74 | 75 | 76 | G:\碩論備份\20180309\LabelingTool\LabelingTool\for_testing\splash.png 77 | G:\碩論備份\20180309\LabelingTool\LabelingTool\for_testing\LabelingTool.exe 78 | G:\碩論備份\20180309\LabelingTool\LabelingTool\for_testing\readme.txt 79 | 80 | 81 | 82 | C:\Program Files\MATLAB\R2017a 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | true 92 | 93 | 94 | 95 | 96 | true 97 | 98 | 99 | 100 | 101 | true 102 | 103 | 104 | 105 | 106 | true 107 | 108 | 109 | 110 | 111 | false 112 | false 113 | true 114 | false 115 | false 116 | false 117 | false 118 | false 119 | 6.2 120 | false 121 | true 122 | win64 123 | true 124 | 125 | 126 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/PackagingLog.html: -------------------------------------------------------------------------------- 1 | 
 2 | mcc -o LabelingTool -W WinMain:LabelingTool -T link:exe -d G:\碩論備份\20180309\LabelingTool\LabelingTool\for_testing -v G:\碩論備份\20180309\LabelingTool\LabelingTool.m -a G:\碩論備份\20180309\LabelingTool\LabelingTool.fig -r 'C:\Program Files\MATLAB\R2017a\toolbox\compiler\Resources\default_icon.ico' 
 3 | Compiler version: 6.4 (R2017a)
 4 | 
 5 | Dependency analysis by REQUIREMENTS.
 6 | 
 7 | Parsing file "G:\碩論備份\20180309\LabelingTool\LabelingTool.m"
 8 | 	(Referenced from: "Compiler Command Line").
 9 | Deleting 2 temporary MEX authorization files.
10 | Removing: 'C:\Users\Andersen\AppData\Local\Temp\b022-660e-5b93-d507_4836.auth'.
11 | Removing: 'C:\Users\Andersen\AppData\Local\Temp\e195-2d04-f447-9530_4836.auth'.
12 | Generating file "G:\碩論備份\20180309\LabelingTool\LabelingTool\for_testing\readme.txt".
13 | Packaging...
14 | Creating the bundle...
15 | Web based installer created at G:\碩論備份\20180309\LabelingTool\LabelingTool\for_redistribution\MyAppInstaller_web.exe.
16 | Packaging complete.
17 | Elapsed packaging time was: 1 minutes and 0 seconds.
18 |
19 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/Video_to_Frame.py: -------------------------------------------------------------------------------- 1 | # Program To Read video 2 | # and Extract Frames 3 | import cv2 4 | import time 5 | 6 | # Function to extract frames 7 | def FrameCapture(path): 8 | # Path to video file 9 | vidObj = cv2.VideoCapture(path) 10 | 11 | # Used as counter variable 12 | count = 0 13 | frame_rate = 29.97 14 | prev = 0 15 | # checks whether frames were extracted 16 | success = 1 17 | 18 | while success: 19 | # vidObj object calls read 20 | # function extract frames 21 | 22 | time_elapsed = time.time() - prev 23 | success, image = vidObj.read() 24 | 25 | if time_elapsed > 1. / frame_rate: 26 | prev = time.time() 27 | cv2.imwrite("C:/Users/rswai/Google_Drive/AIML/Projects/TrackNetMirror-master/Data/images/frame_%d.jpg" % count, image) 28 | 29 | count += 1 30 | 31 | # Driver Code 32 | if __name__ == '__main__': 33 | # Calling the function 34 | FrameCapture(r"C:\Users\rswai\Google_Drive\AIML\Projects\TrackNetMirror-master\Data\vids\vid1_1.mp4") 35 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_redistribution/MyAppInstaller_web.exe: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool/for_redistribution/MyAppInstaller_web.exe -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_redistribution_files_only/LabelingTool.exe: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool/for_redistribution_files_only/LabelingTool.exe -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_redistribution_files_only/default_icon.ico: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool/for_redistribution_files_only/default_icon.ico -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_redistribution_files_only/readme.txt: -------------------------------------------------------------------------------- 1 | MATLAB Compiler 2 | 3 | 1. Prerequisites for Deployment 4 | 5 | . Verify the MATLAB Runtime is installed and ensure you 6 | have installed version 9.2 (R2017a). 7 | 8 | . If the MATLAB Runtime is not installed, do the following: 9 | (1) enter 10 | 11 | >>mcrinstaller 12 | 13 | at MATLAB prompt. The MCRINSTALLER command displays the 14 | location of the MATLAB Runtime installer. 15 | 16 | (2) run the MATLAB Runtime installer. 17 | 18 | Or download the Windows 64-bit version of the MATLAB Runtime for R2017a 19 | from the MathWorks Web site by navigating to 20 | 21 | http://www.mathworks.com/products/compiler/mcr/index.html 22 | 23 | 24 | For more information about the MATLAB Runtime and the MATLAB Runtime installer, see 25 | Package and Distribute in the MATLAB Compiler documentation 26 | in the MathWorks Documentation Center. 27 | 28 | 29 | NOTE: You will need administrator rights to run MCRInstaller. 30 | 31 | 32 | 2. Files to Deploy and Package 33 | 34 | Files to package for Standalone 35 | ================================ 36 | -LabelingTool.exe 37 | -MCRInstaller.exe 38 | -if end users are unable to download the MATLAB Runtime using the above 39 | link, include it when building your component by clicking 40 | the "Runtime downloaded from web" link in the Deployment Tool 41 | -This readme file 42 | 43 | 3. Definitions 44 | 45 | For information on deployment terminology, go to 46 | http://www.mathworks.com/help. Select MATLAB Compiler > 47 | Getting Started > About Application Deployment > 48 | Deployment Product Terms in the MathWorks Documentation 49 | Center. 50 | 51 | 52 | 53 | 54 | 55 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_redistribution_files_only/splash.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool/for_redistribution_files_only/splash.png -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_testing/LabelingTool.exe: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool/for_testing/LabelingTool.exe -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_testing/mccExcludedFiles.log: -------------------------------------------------------------------------------- 1 | The List of Excluded Files 2 | Excluded files Exclusion Message ID Reason For Exclusion Exclusion Rule 3 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_testing/readme.txt: -------------------------------------------------------------------------------- 1 | MATLAB Compiler 2 | 3 | 1. Prerequisites for Deployment 4 | 5 | . Verify the MATLAB Runtime is installed and ensure you 6 | have installed version 9.2 (R2017a). 7 | 8 | . If the MATLAB Runtime is not installed, do the following: 9 | (1) enter 10 | 11 | >>mcrinstaller 12 | 13 | at MATLAB prompt. The MCRINSTALLER command displays the 14 | location of the MATLAB Runtime installer. 15 | 16 | (2) run the MATLAB Runtime installer. 17 | 18 | Or download the Windows 64-bit version of the MATLAB Runtime for R2017a 19 | from the MathWorks Web site by navigating to 20 | 21 | http://www.mathworks.com/products/compiler/mcr/index.html 22 | 23 | 24 | For more information about the MATLAB Runtime and the MATLAB Runtime installer, see 25 | Package and Distribute in the MATLAB Compiler documentation 26 | in the MathWorks Documentation Center. 27 | 28 | 29 | NOTE: You will need administrator rights to run MCRInstaller. 30 | 31 | 32 | 2. Files to Deploy and Package 33 | 34 | Files to package for Standalone 35 | ================================ 36 | -LabelingTool.exe 37 | -MCRInstaller.exe 38 | -if end users are unable to download the MATLAB Runtime using the above 39 | link, include it when building your component by clicking 40 | the "Runtime downloaded from web" link in the Deployment Tool 41 | -This readme file 42 | 43 | 3. Definitions 44 | 45 | For information on deployment terminology, go to 46 | http://www.mathworks.com/help. Select MATLAB Compiler > 47 | Getting Started > About Application Deployment > 48 | Deployment Product Terms in the MathWorks Documentation 49 | Center. 50 | 51 | 52 | 53 | 54 | 55 | -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_testing/requiredMCRProducts.txt: -------------------------------------------------------------------------------- 1 | 35000 35010 -------------------------------------------------------------------------------- /LabelingTool/LabelingTool/for_testing/splash.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/LabelingTool/LabelingTool/for_testing/splash.png -------------------------------------------------------------------------------- /LabelingTool/README.md: -------------------------------------------------------------------------------- 1 | # README for labeling tool 2 | 3 | The excutable file is under directory LabelingTool/for_redistribution_files_only, named LabelingTool.exe 4 | 5 | You need to check if the MATLAB Runtime is installed before you run the program. 6 | 7 | ## How to use it? 8 | 9 | 1. Open the LabelingTool.exe and you will see a window has a main region for image. 10 | 2. Click the "Open" button at up-left corner to open the image source, choose and directory for all images. The image will show in main region in the window. 11 | 3. Move the cursor on the target and click to label it. 12 | 4. The side bar on the right part are the visibility and status of the target: 13 | * Visibility is divided into 4 parts: No Ball, Easy Identification, Hard Identification, Occluded Ball 14 | * Status has three circumstances: Flying, Hit, Bouncing 15 | 5. Using right arrow key and left arrow key on keyborad can move forward and backward image 16 | 6. The text above the main region is noticing user if the image is labeled or not: 17 | * If it is blue, means already labeled 18 | * If it is red, means not label yet 19 | 7. After labeled all images, click the "Output" button at right down corner to output the labeling file. The labeling file will save as csv file, named "Label.csv" under same directory of images. -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # TrackNet 2 | Heatmap based high speed tiny sport objects tracking based on TrackNet 3 | 4 | ![TrackNet](/TrackNet.gif) 5 | 6 | 7 | ## Usage 8 | 9 | ### Installation 10 | 11 | ```bash 12 | pip install -r requirements.txt 13 | ``` 14 | 15 | 16 | ### Inference 17 | 18 | ```bash 19 | # video file with output 20 | cd Code_Baddy 21 | python predict_video.py --save_weights_path=weights/model_baddy.h5 --input_video_path=path_to_file.mp4 --output_video_path="tracked.mp4" --n_classes=256 22 | ``` 23 | 24 | ### Labeling Tool: Make your own dataset 25 | See the readme file in the LabelingTool directory 26 | 27 | ### Training 28 | 29 | 1. Use the TrackNet_Python.ipynb to create heatmap as Ground Truth, and save heatmap as JPG file. (First Cell) 30 | 31 | 2. Create the training and testing csv files (Second cell) 32 | Change the folder path to your custom paths for 1. and 2. 33 | 3. Copy the training.csv and testing.csv file to Code_Baddy/Code_Custom folder 34 | 4. After have training images and ground truth, we can start to train the TrackNet model II (Three frame Input) 35 | * Open command line 36 | * Change directory to Code_Baddy/Code_Custom 37 | * Using following command as example, you may need to change the command: 38 | 39 | ```bash 40 | # Train with custom dataset 41 | python train.py --save_weights_path=weights/model.h5 --training_images_name="training.csv" --epochs=500 --n_classes=256 -- input_height=360 --input_width=640 --load_weights=2 --step_per_epochs=200 --batch_size=2 42 | ``` 43 | 44 | * Trained model weight will be save in weights/model.h5 45 | 46 | * Detailed explanation 47 | --save_weights_path: Save the weight path 48 | --training_images_name: Training images csv file path, training.csv 49 | --epochs: Epochs be set as 500 in this work 50 | --n_classes: Last layer classes, since the output value of TrackNet is between 0-255, the last layer depth be set as 256 51 | --input_height: Input height be resize as 360 in this work 52 | --input_width: Input width be resize as 640 in this work 53 | --load_weights: If you want to retrain the weights from previous weight, give the number of weight in weights/model. If not, delete it. 54 | --step_per_epochs: Step per Epochs be set as 200 in this work 55 | -batch_size: Batch size be set as 2 in this work 56 | 57 | ## References 58 | 59 | GitHub : [TrackNet](https://nol.cs.nctu.edu.tw:234/open-source/TrackNet/tree/master) 60 | 61 | Paper : [TrackNet: A Deep Learning Network for Tracking High-speed and Tiny Objects in Sports Applications](https://ieeexplore.ieee.org/abstract/document/8909871/authors#authors) 62 | -------------------------------------------------------------------------------- /Resources/README.md: -------------------------------------------------------------------------------- 1 | # TrackNet: Tennis Ball Tracking from Broadcast Video by Deep Learning Networks 2 | 3 | ## First, you have to install cuda, cudnn and tensorflow, tutorial: 4 | https://medium.com/@zhanwenchen/install-cuda-and-cudnn-for-tensorflow-gpu-on-ubuntu-79306e4ac04e 5 | 6 | ## Second, install some python library with pip: 7 | * sudo pip install numpy 8 | * sudo pip install matplotlib 9 | * sudo pip install pillow 10 | * sudo pip install keras 11 | * sudo pip install opencv-python 12 | * sudo pip install pydot 13 | * sudo pip install h5py 14 | * sudo apt-get install graphviz 15 | 16 | 17 | 18 | ## How to train a new TrackNet I weight? 19 | 1. Create heatmap as Ground Truth, and save heatmap as JPG file 20 | Code be save in TrackNet_Python.ipynb (first part), you may need to change the folder path in python code 21 | 2. The training file name and testing file name of TrackNet must be output as csv file 22 | Code be save in TrackNet_Python.ipynb (second part), you also need to change the folder path in code 23 | 3. Copy the training_model1.csv file and testing_model1.csv file to TrackNet_One_Frames_Input folder 24 | 4. After have training images and ground truth, we can start to train the TrackNet model I 25 | * Open command line 26 | * Change directory to TrackNet_One_Frames_Input folder 27 | * Using following command as example, you may need to change the command: 28 | 29 | > python train.py --save_weights_path=weights/model --training_images_name="training_model1.csv" --epochs=500 --n_classes=256 --input_height=360 --input_width=640 --load_weights=1 --step_per_epochs=200 --batch_size=2 30 | * Trained model weight will be save in weights/model.0 31 | 32 | * Detailed explanation 33 | --save_weights_path: Save the weight path 34 | --training_images_name: Training images csv file path, training_model1.csv 35 | --epochs: Epochs be set as 500 in this work 36 | --n_classes: Last layer classes, since the output value of TrackNet is between 0-255, the last layer depth be set as 256 37 | --input_height: Input height be resize as 360 in this work 38 | --input_width: Input width be resize as 640 in this work 39 | --load_weights: If you want to retrain the weights from previous weight, give the number of weight in weights/model. If not, delete it. 40 | --step_per_epochs: Step per Epochs be set as 200 in this work 41 | -batch_size: Batch size be set as 2 in this work 42 | 43 | 44 | ## How to train a new TrackNet II weight? 45 | 1. Create heatmap as Ground Truth, and save heatmap as JPG file 46 | Code be saved in TrackNet_Python.ipynb (first part), you may need to change the folder path in python code 47 | 2. The training file name and testing file name of TrackNet must be output as csv file 48 | Code be save in TrackNet_Python.ipynb (third part), you also need to change the folder path in code 49 | 3. Copy the training_model2.csv file and testing_model2.csv file to TrackNet_Three_Frames_Input folder 50 | 4. After have training images and ground truth, we can start to train the TrackNet model II 51 | * Open command line 52 | * Change directory to TrackNet_Three_Frames_Input folder 53 | * Using following command as example, you may need to change the command: 54 | 55 | > python train.py --save_weights_path=weights/model --training_images_name="training_model2.csv" --epochs=500 --n_classes=256 --input_height=360 --input_width=640 --load_weights=2 --step_per_epochs=200 --batch_size=2 56 | * Trained model weight will be save in weights/model.0 57 | 58 | * Detailed explanation 59 | --save_weights_path: Save the weight path 60 | --training_images_name: Training images csv file path, training_model2.csv 61 | --epochs: Epochs be set as 500 in this work 62 | --n_classes: Last layer classes, since the output value of TrackNet is between 0-255, the last layer depth be set as 256 63 | --input_height: Input height be resize as 360 in this work 64 | --input_width: Input width be resize as 640 in this work 65 | --load_weights: If you want to retrain the weights from previous weight, give the number of weight in weights/model. If not, delete it. 66 | --step_per_epochs: Step per Epochs be set as 200 in this work 67 | -batch_size: Batch size be set as 2 in this work 68 | 69 | ## How to train a new TrackNet II' weight? 70 | 1. Create heatmap as Ground Truth, and save heatmap as JPG file 71 | Code be save in TrackNet_Python.ipynb (first part), you may need to change the folder path in python code 72 | 2. The training file name and testing file name of TrackNet must be output as csv file 73 | Code be save in TrackNet_Python.ipynb (third part, fourth part amd fifth part), you also need to change the folder path in code 74 | 3. Copy the training_model3.csv file to TrackNet_Three_Frames_Input folder 75 | 4. After have training images and ground truth, we can start to train the TrackNet model II' 76 | * Open command line 77 | * Change directory to TrackNet_Three_Frames_Input folder 78 | * Using following command as example, you may need to change the command: 79 | 80 | > python train.py --save_weights_path=weights/model --training_images_name="training_model3.csv" --epochs=500 --n_classes=256 --input_height=360 --input_width=640 --load_weights=2 --step_per_epochs=200 --batch_size=2 81 | * Trained model weight will be save in weights/model.0 82 | 83 | * Detailed explanation 84 | --save_weights_path: Save the weight path 85 | --training_images_name: Training images csv file path, training_model3.csv 86 | --epochs: Epochs be set as 500 in this work 87 | --n_classes: Last layer classes, since the output value of TrackNet is between 0-255, the last layer depth be set as 256 88 | --input_height: Input height be resize as 360 in this work 89 | --input_width: Input width be resize as 640 in this work 90 | --load_weights: If you want to retrain the weights from previous weight, give the number of weight in weights/model. If not, delete it. 91 | --step_per_epochs: Step per Epochs be set as 200 in this work 92 | -batch_size: Batch size be set as 2 in this work 93 | 94 | ## How to output all of heatmap predictions? 95 | 1. Open command line 96 | 2. Change directory to TrackNet folder (TrackNet_Three_Frames_Input or TrackNet_One_Frames_Input) 97 | 3. Using following command as example, you may need to change the command: 98 | 99 | python predict.py --save_weights_path=weights/model.2 --test_images="/media/andersen/D/Thesis/Dataset/Clip" --output_path="/media/andersen/D/Thesis/Prediction/Model2/Clip" --n_classes=256 --input_height=360 --input_width=640 --output_height=720 --output_width=1280 100 | 101 | * Detailed explanation 102 | --save_weights_path: which model weight need to be loaded 103 | --test_images: testing images path 104 | --output_path: output heatmap path 105 | --n_classes: in this work depth be set as 256 106 | --input_height: Input height be resize as 360 in this work 107 | --input_width: Input width be resize as 360 in this work 108 | --output_height: resize the heatmap height, output height be set as 720 in this work 109 | -output_width: resize the heatmap width,output width be set as 1280 in this work 110 | 111 | 112 | 113 | ## How to use TrackNet predict video? 114 | 1. Open command line 115 | 2. Change directory to TrackNet folder (TrackNet_Three_Frames_Input or TrackNet_One_Frames_Input) 116 | 3. using following command as example, you may need to change the command: 117 | 118 | python predict_video.py --save_weights_path=weights/model.3 --input_video_path="/media/andersen/D/Test/Clip1.mp4" --output_video_path="/media/andersen/D/Test/Clip1_TrackNet.mp4" --n_classes=256 119 | 120 | * Detailed explanation 121 | --save_weights_path: which model weight need to be loaded 122 | --input_video_path: Input video path 123 | --output_video_path: Output video path, if not, the video will be save in the same path of input video 124 | --n_classes: In this work depth be set as 256 125 | 126 | 127 | 128 | ## TrackNet trained weights: 129 | * TrackNet model I >> TrackNet_One_Frames_Input/weights.model.1 130 | * TrackNet model II >> TrackNet_Three_Frames_Input/weights.model.2 131 | * TrackNet model II' >> TrackNet_Three_Frames_Input/weights.model.3 132 | 133 | 134 | 135 | ## Labeling Tool: Make your own dataset 136 | See the readme file in the LabelingTool directory -------------------------------------------------------------------------------- /Resources/TrackNet.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Resources/TrackNet.pdf -------------------------------------------------------------------------------- /Resources/TrackNet.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Resources/TrackNet.png -------------------------------------------------------------------------------- /Resources/coachAI.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/Resources/coachAI.pdf -------------------------------------------------------------------------------- /TrackNet.gif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/weekenddeeplearning/TrackNet/2836c388234fce6dfc967f8c694d0970928258ea/TrackNet.gif -------------------------------------------------------------------------------- /Utils/frame_to_vid.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import os 3 | 4 | image_folder = r'C:\Users\rswai\Google_Drive\AIML\Projects\TrackNetMirror-master\Data\Tennis\game10\Clip12' 5 | video_name = 'output.mp4' 6 | 7 | images = [img for img in os.listdir(image_folder) if img.endswith(".jpg")] 8 | frame = cv2.imread(os.path.join(image_folder, images[0])) 9 | height, width, layers = frame.shape 10 | 11 | # fourcc = cv2.cv.CV_FOURCC('MP4V') 12 | fourcc = cv2.VideoWriter_fourcc(*'mp4v') 13 | 14 | video = cv2.VideoWriter(video_name, fourcc, 30, (width,height)) 15 | 16 | for image in images: 17 | video.write(cv2.imread(os.path.join(image_folder, image))) 18 | 19 | cv2.destroyAllWindows() 20 | video.release() -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | argparse 2 | numpy 3 | matplotlib 4 | pillow 5 | h5py 6 | pydot 7 | keras 8 | tensorflow-gpu 9 | opencv-python 10 | --------------------------------------------------------------------------------