├── .idea ├── .gitignore ├── encodings.xml ├── libraries │ └── R_User_Library.xml ├── misc.xml ├── modules.xml ├── vcs.xml └── 基于Paddlepaddle的黑白影片色彩重建.iml ├── Prior ├── Clear.py └── imageCrop.py ├── README.md ├── conversion.py ├── gray2color.py ├── main.py ├── model └── gray2color.inference.model │ ├── batch_norm_0.b_0 │ ├── batch_norm_0.w_0 │ ├── batch_norm_0.w_1 │ ├── batch_norm_0.w_2 │ ├── batch_norm_1.b_0 │ ├── batch_norm_1.w_0 │ ├── batch_norm_1.w_1 │ ├── batch_norm_1.w_2 │ ├── batch_norm_10.b_0 │ ├── batch_norm_10.w_0 │ ├── batch_norm_10.w_1 │ ├── batch_norm_10.w_2 │ ├── batch_norm_11.b_0 │ ├── batch_norm_11.w_0 │ ├── batch_norm_11.w_1 │ ├── batch_norm_11.w_2 │ ├── batch_norm_12.b_0 │ ├── batch_norm_12.w_0 │ ├── batch_norm_12.w_1 │ ├── batch_norm_12.w_2 │ ├── batch_norm_13.b_0 │ ├── batch_norm_13.w_0 │ ├── batch_norm_13.w_1 │ ├── batch_norm_13.w_2 │ ├── batch_norm_14.b_0 │ ├── batch_norm_14.w_0 │ ├── batch_norm_14.w_1 │ ├── batch_norm_14.w_2 │ ├── batch_norm_2.b_0 │ ├── batch_norm_2.w_0 │ ├── batch_norm_2.w_1 │ ├── batch_norm_2.w_2 │ ├── batch_norm_3.b_0 │ ├── batch_norm_3.w_0 │ ├── batch_norm_3.w_1 │ ├── batch_norm_3.w_2 │ ├── batch_norm_4.b_0 │ ├── batch_norm_4.w_0 │ ├── batch_norm_4.w_1 │ ├── batch_norm_4.w_2 │ ├── batch_norm_5.b_0 │ ├── batch_norm_5.w_0 │ ├── batch_norm_5.w_1 │ ├── batch_norm_5.w_2 │ ├── batch_norm_6.b_0 │ ├── batch_norm_6.w_0 │ ├── batch_norm_6.w_1 │ ├── batch_norm_6.w_2 │ ├── batch_norm_7.b_0 │ ├── batch_norm_7.w_0 │ ├── batch_norm_7.w_1 │ ├── batch_norm_7.w_2 │ ├── batch_norm_8.b_0 │ ├── batch_norm_8.w_0 │ ├── batch_norm_8.w_1 │ ├── batch_norm_8.w_2 │ ├── batch_norm_9.b_0 │ ├── batch_norm_9.w_0 │ ├── batch_norm_9.w_1 │ ├── batch_norm_9.w_2 │ ├── conv2d_0.b_0 │ ├── conv2d_0.w_0 │ ├── conv2d_1.b_0 │ ├── conv2d_1.w_0 │ ├── conv2d_10.b_0 │ ├── conv2d_10.w_0 │ ├── conv2d_11.b_0 │ ├── conv2d_11.w_0 │ ├── conv2d_12.b_0 │ ├── conv2d_12.w_0 │ ├── conv2d_13.b_0 │ ├── conv2d_13.w_0 │ ├── conv2d_14.b_0 │ ├── conv2d_14.w_0 │ ├── conv2d_2.b_0 │ ├── conv2d_2.w_0 │ ├── conv2d_3.b_0 │ ├── conv2d_3.w_0 │ ├── conv2d_4.b_0 │ ├── conv2d_4.w_0 │ ├── conv2d_5.b_0 │ ├── conv2d_5.w_0 │ ├── conv2d_6.b_0 │ ├── conv2d_6.w_0 │ ├── conv2d_7.b_0 │ ├── conv2d_7.w_0 │ ├── conv2d_8.b_0 │ ├── conv2d_8.w_0 │ ├── conv2d_9.b_0 │ ├── conv2d_9.w_0 │ ├── conv2d_transpose_0.w_0 │ └── conv2d_transpose_1.w_0 ├── resouse └── logo.png └── train ├── ResNet.py ├── Utils.py └── fileReader.py /.idea/.gitignore: -------------------------------------------------------------------------------- 1 | # Default ignored files 2 | /workspace.xml -------------------------------------------------------------------------------- /.idea/encodings.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | -------------------------------------------------------------------------------- /.idea/libraries/R_User_Library.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | -------------------------------------------------------------------------------- /.idea/misc.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | -------------------------------------------------------------------------------- /.idea/modules.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/vcs.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | -------------------------------------------------------------------------------- /.idea/基于Paddlepaddle的黑白影片色彩重建.iml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | -------------------------------------------------------------------------------- /Prior/Clear.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import os 3 | import imghdr 4 | from PIL import Image 5 | import threading 6 | 7 | 8 | '''多线程将数据集中单通道图删除''' 9 | def cutArray(l, num): 10 | avg = len(l) / float(num) 11 | o = [] 12 | last = 0.0 13 | 14 | while last < len(l): 15 | o.append(l[int(last):int(last + avg)]) 16 | last += avg 17 | 18 | return o 19 | 20 | def deleteErrorImage(path,image_dir): 21 | count = 0 22 | for file in image_dir: 23 | try: 24 | image = os.path.join(path,file) 25 | image_type = imghdr.what(image) 26 | if image_type is not 'jpeg': 27 | os.remove(image) 28 | count = count + 1 29 | 30 | img = np.array(Image.open(image)) 31 | if len(img.shape) is 2: 32 | os.remove(image) 33 | count = count + 1 34 | except Exception as e: 35 | print(e) 36 | print('done!') 37 | print('已删除数量：' + str(count)) 38 | 39 | class thread(threading.Thread): 40 | def __init__(self, threadID, path, files): 41 | threading.Thread.__init__(self) 42 | self.threadID = threadID 43 | self.path = path 44 | self.files = files 45 | def run(self): 46 | deleteErrorImage(self.path,self.files) 47 | 48 | if __name__ == '__main__': 49 | path = './work/train/' 50 | files = os.listdir(path) 51 | files = cutArray(files,8) 52 | t1 = threading.Thread(target=deleteErrorImage,args=(path,files[0])) 53 | t2 = threading.Thread(target=deleteErrorImage,args=(path,files[1])) 54 | t3 = threading.Thread(target=deleteErrorImage,args=(path,files[2])) 55 | t4 = threading.Thread(target=deleteErrorImage,args=(path,files[3])) 56 | t5 = threading.Thread(target=deleteErrorImage,args=(path,files[4])) 57 | t6 = threading.Thread(target=deleteErrorImage,args=(path,files[5])) 58 | t7 = threading.Thread(target=deleteErrorImage,args=(path,files[6])) 59 | t8 = threading.Thread(target=deleteErrorImage,args=(path,files[7])) 60 | threadList = [] 61 | threadList.append(t1) 62 | threadList.append(t2) 63 | threadList.append(t3) 64 | threadList.append(t4) 65 | threadList.append(t5) 66 | threadList.append(t6) 67 | threadList.append(t7) 68 | threadList.append(t8) 69 | for t in threadList: 70 | t.setDaemon(True) 71 | t.start() 72 | t.join() -------------------------------------------------------------------------------- /Prior/imageCrop.py: -------------------------------------------------------------------------------- 1 | from PIL import Image 2 | import os.path 3 | import os 4 | import threading 5 | from PIL import ImageFile 6 | ImageFile.LOAD_TRUNCATED_IMAGES = True 7 | 8 | '''多线程将图片缩放后再裁切到256*256分辨率''' 9 | w = 256 10 | h = 256 11 | 12 | def cutArray(l, num): 13 | avg = len(l) / float(num) 14 | o = [] 15 | last = 0.0 16 | 17 | while last < len(l): 18 | o.append(l[int(last):int(last + avg)]) 19 | last += avg 20 | 21 | return o 22 | 23 | def convertjpg(jpgfile,outdir,width=w,height=h): 24 | img=Image.open(jpgfile) 25 | (l,h) = img.size 26 | rate = min(l,h) / width 27 | try: 28 | img = img.resize((int(l // rate),int(h // rate)),Image.BILINEAR) 29 | (l,h) = img.size 30 | lstart = (l - width)//2 31 | hstart = (h - height)//2 32 | img = img.crop((lstart,hstart,lstart + width,hstart + height)) 33 | img.save(os.path.join(outdir,os.path.basename(jpgfile))) 34 | except Exception as e: 35 | print(e) 36 | 37 | class thread(threading.Thread): 38 | def __init__(self, threadID, inpath, outpath, files): 39 | threading.Thread.__init__(self) 40 | self.threadID = threadID 41 | self.inpath = inpath 42 | self.outpath = outpath 43 | self.files = files 44 | def run(self): 45 | count = 0 46 | try: 47 | for file in self.files: 48 | convertjpg(self.inpath + file,self.outpath) 49 | count = count + 1 50 | except Exception as e: 51 | print(e) 52 | print('已处理图片数量：' + str(count)) 53 | 54 | if __name__ == '__main__': 55 | inpath = './work/train/' 56 | outpath = './work/train/' 57 | files = os.listdir(inpath) 58 | # for file in files: 59 | # convertjpg(path + file,path) 60 | files = cutArray(files,8) 61 | T1 = thread(1, inpath, outpath, files[0]) 62 | T2 = thread(2, inpath, outpath, files[1]) 63 | T3 = thread(3, inpath, outpath, files[2]) 64 | T4 = thread(4, inpath, outpath, files[3]) 65 | T5 = thread(5, inpath, outpath, files[4]) 66 | T6 = thread(6, inpath, outpath, files[5]) 67 | T7 = thread(7, inpath, outpath, files[6]) 68 | T8 = thread(8, inpath, outpath, files[7]) 69 | 70 | T1.start() 71 | T2.start() 72 | T3.start() 73 | T4.start() 74 | T5.start() 75 | T6.start() 76 | T7.start() 77 | T8.start() 78 | 79 | T1.join() 80 | T2.join() 81 | T3.join() 82 | T4.join() 83 | T5.join() 84 | T6.join() 85 | T7.join() 86 | T8.join() -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ![image](resouse/logo.png) 2 | 3 | **通过黑白图片的情景语义找到颜色和结构纹理特征的映射,将黑白影片彩色化** 4 | 5 | | 脚本需要配合命令行工具ffmpeg | 6 | | -------------------------- | 7 | ## 项目文件结构 8 | 9 | + cache 视频转换缓存目录 10 | * audio 原视频的音频文件 11 | * input_img 原视频分解成每帧图像目录(黑白图片) 12 | * output_img 将input_img文件夹下图片处理后生成对应的图片目录(彩色图片) 13 | + data 训练集目录 14 | * test 测试集 15 | * train 训练集 16 | + model 模型存放目录 17 | + movie 视频文件(*.mp4)文件存放目录 18 | + prior 预处理目录 19 | * Clear.py 清理训练集和测试集中的损坏文件和不是JPEG格式的图片 20 | * getExperience.py 预处理ImageNet数据集得到Q空间上的经验分布 21 | * getWeight.py 将经验分布结合高斯函数，经过处理得到经验分布的权值 22 | * imageCrop.py 将图片进行缩放后再裁剪到256*256像素 23 | * Experience.npy 运行getExperience.py生成经验分布的Numpy矩阵文件 24 | * Q.npy 人为划定的Q空间对应的Numpy矩阵文件 25 | * Weight.npy 运行getWeight.py生成经验分布的Numpy矩阵文件 26 | + train 训练脚本目录 27 | * fileReader.npy 从指定文件夹读取文件的Reader，并做预处理 28 | * train.npy 训练脚本 29 | + conversion.py 将movie目录读取input.mp4文件并分解成每帧图片存放在cache/input_img/,音频文件存放在cache/audio/ 30 | + gray2color.py 将输入图片经过模型处理再转换为RGB图片并储存 31 | 32 | ## 转换图片/视频 33 | 先进入paddle环境 34 | ```bash 35 | cd <项目路径> 36 | source activate 37 | ``` 38 | 转换图片 39 | ```bash 40 | Python main.py -img_in <需要转换图片路径> -save <保存位置> 41 | ``` 42 | 转换视频 43 | ```bash 44 | Python main.py -movie_in <需要转换视频路径> -save <保存位置> 45 | ``` 46 | # PaddleColorization-黑白照片着色 47 | 48 | > 将黑白照片着色是不是一件神奇的事情？ 49 | 50 | > 本项目将带领你一步一步学习将黑白图片甚至黑白影片的彩色化 51 | 52 | --- 53 | 54 | ## 开启着色之旅！！！ 55 | 56 | ### 先来看看成品 57 | 58 | ![](https://ai-studio-static-online.cdn.bcebos.com/bae1ccceaa4b4c34af7a3f1667547380caabb5fa54d749ef981150601a587935) 59 | 60 | > 欢迎大家fork学习~有任何问题欢迎在评论区留言互相交流哦 61 | 62 | > 这里一点小小的宣传，我感兴趣的领域包括迁移学习、生成对抗网络。欢迎交流关注。[来AI Studio互粉吧~等你哦~ ](https://aistudio.baidu.com/aistudio/personalcenter/thirdview/56447) 63 | 64 | 65 | # 1 项目简介 66 | 67 | 本项目基于paddlepaddle，结合残差网络（ResNet）,通过监督学习的方式，训练模型将黑白图片转换为彩色图片 68 | 69 | --- 70 | 71 | ### 1.1 残差网络（ResNet） 72 | 73 | ### 1.1.1 背景介绍 74 | 75 | ResNet(Residual Network) [15] 是2015年ImageNet图像分类、图像物体定位和图像物体检测比赛的冠军。针对随着网络训练加深导致准确度下降的问题，ResNet提出了残差学习方法来减轻训练深层网络的困难。在已有设计思路(BN, 小卷积核，全卷积网络)的基础上，引入了残差模块。每个残差模块包含两条路径，其中一条路径是输入特征的直连通路，另一条路径对该特征做两到三次卷积操作得到该特征的残差，最后再将两条路径上的特征相加。 76 | 77 | 残差模块如图9所示，左边是基本模块连接方式，由两个输出通道数相同的3x3卷积组成。右边是瓶颈模块(Bottleneck)连接方式，之所以称为瓶颈，是因为上面的1x1卷积用来降维(图示例即256->64)，下面的1x1卷积用来升维(图示例即64->256)，这样中间3x3卷积的输入和输出通道数都较小(图示例即64->64)。 78 | 79 | ![](https://ai-studio-static-online.cdn.bcebos.com/7ede3132804549228b5c4a729d90e6b25821272dd9e74e41a95d3363f9e06c0e) 80 | 81 | ### 1.2 项目设计思路及主要解决问题 82 | 83 | > * 设计思路：通过训练网络对大量样本的学习得到经验分布（例如天空永远是蓝色的，草永远是绿色的），通过经验分布推得黑白图像上各部分合理的颜色 84 | 85 | > * 主要解决问题：大量物体颜色并不是固定的也就是物体颜色具有多模态性（例如：苹果可以是红色也可以是绿色和黄色）。通常使用均方差作为损失函数会让具有颜色多模态属性的物体趋于寻找一个“平均”的颜色（通常为淡黄色）导致着色后的图片饱和度不高。 86 | 87 | --- 88 | 89 | ### 1.3 本文主要特征 90 | 91 | * 将Adam优化器beta1参数设置为0.8，具体请参考[原论文](https://arxiv.org/abs/1412.6980) 92 | * 将BatchNorm批归一化中momentum参数设置为0.5 93 | * 采用基本模块连接方式 94 | * 为抑制多模态问题，在均方差的基础上重新设计损失函数 95 | 96 | --- 97 | 98 | 损失函数公式如下: 99 | 100 | $Out = 1/n\sum{(input-label)^{2}} + 26.7/(n{\sum{(input - \bar{input})^{2}}})$ 101 | 102 | --- 103 | 104 | ### 1.4 数据集介绍（ImageNet） 105 | 106 | > ImageNet项目是一个用于视觉对象识别软件研究的大型可视化数据库。超过1400万的图像URL被ImageNet手动注释，以指示图片中的对象;在至少一百万个图像中，还提供了边界框。ImageNet包含2万多个类别; [2]一个典型的类别，如“气球”或“草莓”，包含数百个图像。第三方图像URL的注释数据库可以直接从ImageNet免费获得;但是，实际的图像不属于ImageNet。自2010年以来，ImageNet项目每年举办一次软件比赛，即ImageNet大规模视觉识别挑战赛（ILSVRC），软件程序竞相正确分类检测物体和场景。 ImageNet挑战使用了一个“修剪”的1000个非重叠类的列表。2012年在解决ImageNet挑战方面取得了巨大的突破，被广泛认为是2010年的深度学习革命的开始。（来源：百度百科） 107 | 108 | > ImageNet2012介绍： 109 | > 110 | > * Training images (Task 1 & 2). 138GB.（约120万张高清图片，共1000个类别） 111 | > * Validation images (all tasks). 6.3GB. 112 | > * Training bounding box annotations (Task 1 & 2 only). 20MB. 113 | 114 | ### 1.5 LAB颜色空间 115 | 116 | > Lab模式是根据Commission International Eclairage（CIE）在1931年所制定的一种测定颜色的国际标准建立的。于1976年被改进，并且命名的一种色彩模式。Lab颜色模型弥补了RGB和CMYK两种色彩模式的不足。它是一种设备无关的颜色模型，也是一种基于生理特征的颜色模型。 [1] Lab颜色模型由三个要素组成，一个要素是亮度（L），a 和b是两个颜色通道。a包括的颜色是从深绿色（低亮度值）到灰色（中亮度值）再到亮粉红色（高亮度值）；b是从亮蓝色（低亮度值）到灰色（中亮度值）再到黄色（高亮度值）。因此，这种颜色混合后将产生具有明亮效果的色彩。（来源：百度百科） 117 | > ![](https://ai-studio-static-online.cdn.bcebos.com/dd030a2a8d22435e815e385d017bf87190ea4cfb1368491f8e08f428104863d0) 118 | 119 | 120 | 121 | 122 | # 2.项目总结 123 | 124 | 通过循序渐进的方式叙述了项目的过程。 125 | 对于训练结果虽然本项目通过抑制平均化加大离散程度提高了着色的饱和度，但最终结果仍然有较大现实差距，只能对部分场景有比较好的结果，对人造场景（如超市景观等）仍然表现力不足。 126 | 接下来准备进一步去设计损失函数，目的是让网络着色结果足以欺骗人的”直觉感受“，而不是一味地接近真实场景 -------------------------------------------------------------------------------- /conversion.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | from cv2 import VideoWriter,VideoWriter_fourcc 3 | import os 4 | import shutil 5 | import subprocess 6 | 7 | '''本脚本需要配合命令行工具ffmpeg''' 8 | 9 | def video2photo(videoName = './movie/input.mp4'): 10 | '''将视频转换为图片，视频默认读取路径./movie/input.mp4，图片写入路径./cache/input_img/*（从1开始给每一帧图片命名*.jpg) 11 | INPUTS 12 | videoName 视频读取路径 13 | OUTPUTS 14 | None 15 | ''' 16 | video = cv2.VideoCapture(videoName) 17 | n = 1 18 | #清空图片缓存 19 | shutil.rmtree('./cache/input_img') 20 | os.mkdir('./cache/input_img') 21 | #清空音频缓存 22 | shutil.rmtree('./cache/audio') 23 | os.mkdir('./cache/audio') 24 | # 提取音频保存到./cache/audio/ 25 | # 运行终端程序提取音频 26 | command = 'ffmpeg -i ' + videoName + ' -ab 160k -ac 2 -ar 44100 -vn ./cache/audio/audio.wav' 27 | subprocess.call(command, shell=True,stdout= open('/dev/null','w'),stderr=subprocess.STDOUT) 28 | 29 | if video.isOpened(): 30 | rval, frame = video.read() 31 | else: 32 | rval = False 33 | while rval: 34 | try: 35 | rval,frame = video.read() 36 | cv2.imwrite('./cache/input_img/' + str(n) + '.jpg',frame) 37 | n = n + 1 38 | cv2.waitKey(1) 39 | except Exception as e: 40 | break 41 | video.release() 42 | print('[*] 视频转换图片成功...') 43 | 44 | 45 | 46 | 47 | def photo2video(videoName = './cache/output.mp4',fps = 25,imgs = './cache/output_img'): 48 | '''将图片转换为视频，图片读取路径./cache/output_img.mp4，视频默认写入路径./movie/output.mp4，待合成图片默认文件夹路径./cache/output_img 49 | INPUTS 50 | videoName 视频写入路径 51 | fps 视频帧率 52 | imgs 待合成图片文件夹路径 53 | OUTPUTS 54 | None 55 | ''' 56 | #获取图像分辨率 57 | img = cv2.imread(imgs + '/1.jpg') 58 | sp = img.shape 59 | height = sp[0] 60 | width = sp[1] 61 | size = (width,height) 62 | 63 | fourcc = VideoWriter_fourcc(*'mp4v') 64 | videoWriter = VideoWriter(videoName,fourcc,fps,size) 65 | photos = os.listdir(imgs) 66 | for photo in range(len(photos)): 67 | frame = cv2.imread(imgs + '/' + str(photo) + '.jpg') 68 | videoWriter.write(frame) 69 | videoWriter.release() 70 | #合成新生成的视频和音频 71 | if (os.path.exists('./movie/output.mkv')): 72 | os.remove('./movie/output.mkv') 73 | command = 'ffmpeg -i ./movie/output.mp4 -i ./cache/audio/audio.wav -c copy ./movie/output.mkv' 74 | subprocess.call(command,shell=True,stdout= open('/dev/null','w'),stderr=subprocess.STDOUT) 75 | #os.remove('./cache/output.mp4') 76 | print('[*] 生成视频成功，路径为./cache/output.mkv') 77 | 78 | 79 | 80 | 81 | '''test''' 82 | if __name__ == '__main__': 83 | photo2video(imgs='./cache/output_img') -------------------------------------------------------------------------------- /gray2color.py: -------------------------------------------------------------------------------- 1 | import paddle 2 | import numpy as np 3 | from skimage import io,color,transform 4 | from paddle import fluid 5 | import matplotlib.pyplot as plt 6 | import os 7 | import lab2rgb 8 | import Prior.imageCrop as crop 9 | import matplotlib.pyplot as plt 10 | 11 | use_cuda = True 12 | place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace() 13 | exe = fluid.Executor(place) 14 | scope = fluid.core.Scope() 15 | fluid.scope_guard(scope) 16 | [inference_program, feed_target_names, fetch_targets] = fluid.io.load_inference_model(dirname=r'model/gray2color.inference.model',executor=exe,) 17 | 18 | def loadImage(image): 19 | '''读取图片,并转为Lab，并提取出L和ab''' 20 | img = io.imread(image) 21 | lab = np.array(color.rgb2lab(img)).transpose() 22 | l = lab[:1, :, :] 23 | return l.reshape(1,1,256,256).astype('float32') 24 | 25 | 26 | def nd_to_2d( i, axis=0): 27 | '''将N维的矩阵转为2维矩阵 28 | INPUT 29 | i N维矩阵 30 | axis 需要保留的维度 31 | OUTPUT 32 | o 转换的2维矩阵 33 | ''' 34 | n = i.ndim 35 | shapeArray = np.array(i.shape) 36 | diff = np.setdiff1d(np.arange(0, n), np.array(axis)) 37 | p = np.prod(shapeArray[diff]) 38 | ax = np.concatenate((diff, np.array(axis).flatten()), axis=0) 39 | o = i.transpose((ax)) 40 | o = o.reshape(p, shapeArray[axis]) 41 | return o 42 | 43 | def run(input,output): 44 | '''处理图片并存储到相应位置 45 | INPUT 46 | input 输入图片路径 47 | output 输出图片路径 48 | OUTPUT 49 | None 50 | ''' 51 | inference_transpiler_program = inference_program.clone() 52 | # t = fluid.transpiler.InferenceTranspiler() 53 | # t.transpile(inference_transpiler_program, place) 54 | crop.convertjpg(jpgfile=input,outdir=None) 55 | l = loadImage(input) 56 | result = exe.run(inference_program, feed={feed_target_names[0]: (l)}, fetch_list=fetch_targets) 57 | 58 | ab = result[0][0] 59 | l = l[0][0] 60 | img = lab2rgb.lab2rgb(l,ab) 61 | #img = transform.rotate(img,270) 62 | #img = np.fliplr(img) 63 | img = img.astype('float32') 64 | plt.grid(False) 65 | plt.axis('off') 66 | plt.imshow(img) 67 | #plt.show() 68 | plt.savefig(str(output),bbox_inches='tight') 69 | 70 | 71 | if __name__ == '__main__': 72 | # run('/home/redflashing/PycharmProjects/基于Paddlepaddle的黑白影片色彩重建/data/train/n01514859_142.JPEG','/home/redflashing/下载/1.JPEG') 73 | path = './dataset/test/' 74 | files = os.listdir(path) 75 | for file in files: 76 | try: 77 | run(path+file,'output_img/'+file) 78 | print(file + ' done!') 79 | except Exception as e: 80 | print(e) 81 | 82 | 83 | 84 | 85 | -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import os 3 | import skimage.color as color 4 | import matplotlib.pyplot as plt 5 | import conversion 6 | import gray2color 7 | import argparse 8 | 9 | def parse_args(): 10 | parser = argparse.ArgumentParser(description='基于paddlepaddle的黑白影片色彩重建') 11 | parser.add_argument('-img_in',dest='img_in',help='输入需要转换的黑白照片路径', type=str) 12 | parser.add_argument('-save',dest='save',help='需要保存结果的路径', type=str, default='./movie/') 13 | parser.add_argument('-movie_in',dest='movie_in',help='输入需要转换的黑白影片的路径', type=str) 14 | args = parser.parse_args() 15 | return args 16 | 17 | if __name__ == '__main__': 18 | args = parse_args() 19 | if args.img_in is not None and args.movie_in is None: 20 | gray2color.run(input=args.img_in,output=args.save) 21 | print('[*] 图片转换成功，路径为' + args.save) 22 | elif args.img_in is None and args.movie_in is not None: 23 | print('[*] 正在转换中...') 24 | conversion.video2photo(videoName=args.movie_in) 25 | cacheIN = 'cache/input_img/' 26 | cacheOUT = 'cache/output_img/' 27 | imgs = os.listdir(cacheIN) 28 | for img in imgs: 29 | gray2color.run(input=cacheIN + img,output= cacheOUT + img) 30 | conversion.photo2video(videoName=args.save,fps=25) 31 | print('[*] 视频转换成功，路径为' + args.save) 32 | else: 33 | print('[*] 输入有误') 34 | 35 | 36 | -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_0.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_0.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_0.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_0.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_0.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_0.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_0.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_0.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_1.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_1.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_1.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_1.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_1.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_1.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_1.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_1.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_10.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_10.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_10.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_10.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_10.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_10.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_10.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_10.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_11.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_11.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_11.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_11.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_11.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_11.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_11.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_11.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_12.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_12.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_12.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_12.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_12.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_12.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_12.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_12.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_13.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_13.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_13.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_13.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_13.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_13.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_13.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_13.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_14.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_14.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_14.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_14.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_14.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_14.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_14.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_14.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_2.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_2.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_2.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_2.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_2.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_2.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_2.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_2.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_3.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_3.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_3.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_3.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_3.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_3.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_3.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_3.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_4.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_4.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_4.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_4.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_4.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_4.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_4.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_4.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_5.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_5.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_5.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_5.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_5.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_5.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_5.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_5.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_6.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_6.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_6.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_6.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_6.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_6.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_6.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_6.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_7.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_7.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_7.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_7.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_7.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_7.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_7.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_7.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_8.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_8.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_8.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_8.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_8.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_8.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_8.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_8.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_9.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_9.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_9.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_9.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_9.w_1: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_9.w_1 -------------------------------------------------------------------------------- /model/gray2color.inference.model/batch_norm_9.w_2: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/batch_norm_9.w_2 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_0.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_0.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_0.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_0.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_1.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_1.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_1.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_1.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_10.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_10.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_10.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_10.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_11.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_11.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_11.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_11.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_12.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_12.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_12.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_12.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_13.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_13.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_13.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_13.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_14.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_14.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_14.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_14.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_2.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_2.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_2.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_2.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_3.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_3.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_3.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_3.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_4.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_4.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_4.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_4.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_5.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_5.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_5.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_5.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_6.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_6.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_6.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_6.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_7.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_7.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_7.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_7.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_8.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_8.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_8.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_8.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_9.b_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_9.b_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_9.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_9.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_transpose_0.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_transpose_0.w_0 -------------------------------------------------------------------------------- /model/gray2color.inference.model/conv2d_transpose_1.w_0: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/model/gray2color.inference.model/conv2d_transpose_1.w_0 -------------------------------------------------------------------------------- /resouse/logo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Siyou-Li/PaddleColorization/ba3e06e41e09d37d73dbf2cf766b60bcb0e1a46f/resouse/logo.png -------------------------------------------------------------------------------- /train/ResNet.py: -------------------------------------------------------------------------------- 1 | import paddle 2 | import paddle.fluid as fluid 3 | import numpy as np 4 | import sys 5 | import os 6 | sys.path.append('./train') 7 | from skimage import io,color 8 | from train.fileReader import * 9 | import matplotlib.pyplot as plt 10 | import lab2rgb 11 | 12 | Params_dirname = "work/model/gray2color.inference.model" 13 | BATCH_SIZE = 30 14 | EPOCH_NUM = 300 15 | 16 | '''自定义损失函数''' 17 | def createLoss(predict, truth): 18 | '''均方差''' 19 | loss1 = fluid.layers.square_error_cost(predict,truth) 20 | #loss2 = fluid.layers.square_error_cost(predict,fluid.layers.fill_constant(shape=[BATCH_SIZE,2,256,256],value=fluid.layers.mean(predict),dtype='float32')) 21 | cost = fluid.layers.mean(loss1) #+ 16.7 / fluid.layers.mean(loss2) 22 | return cost 23 | 24 | def conv_bn_layer(input, 25 | ch_out, 26 | filter_size, 27 | stride, 28 | padding, 29 | act='relu', 30 | bias_attr=True): 31 | tmp = fluid.layers.conv2d( 32 | input=input, 33 | filter_size=filter_size, 34 | num_filters=ch_out, 35 | stride=stride, 36 | padding=padding, 37 | act=None, 38 | bias_attr=bias_attr) 39 | return fluid.layers.batch_norm(input=tmp,act=act,momentum=0.5) 40 | 41 | 42 | def shortcut(input, ch_in, ch_out, stride): 43 | if ch_in != ch_out: 44 | return conv_bn_layer(input, ch_out, 1, stride, 0, None) 45 | else: 46 | return input 47 | 48 | 49 | def basicblock(input, ch_in, ch_out, stride): 50 | tmp = conv_bn_layer(input, ch_out, 3, stride, 1) 51 | tmp = conv_bn_layer(tmp, ch_out, 3, 1, 1, act=None, bias_attr=True) 52 | short = shortcut(input, ch_in, ch_out, stride) 53 | return fluid.layers.elementwise_add(x=tmp, y=short, act='relu') 54 | 55 | 56 | def layer_warp(block_func, input, ch_in, ch_out, count, stride): 57 | tmp = block_func(input, ch_in, ch_out, stride) 58 | for i in range(1, count): 59 | tmp = block_func(tmp, ch_out, ch_out, 1) 60 | return tmp 61 | 62 | ###反卷积层 63 | def deconv(x, num_filters, filter_size=5, stride=2, dilation=1, padding=2, output_size=None, act=None): 64 | return fluid.layers.conv2d_transpose( 65 | input=x, 66 | num_filters=num_filters, 67 | # 滤波器数量 68 | output_size=output_size, 69 | # 输出图片大小 70 | filter_size=filter_size, 71 | # 滤波器大小 72 | stride=stride, 73 | # 步长 74 | dilation=dilation, 75 | # 膨胀比例大小 76 | padding=padding, 77 | use_cudnn=True, 78 | # 是否使用cudnn内核 79 | act=act 80 | # 激活函数 81 | ) 82 | def bn(x, name=None, act=None,momentum=0.5): 83 | return fluid.layers.batch_norm( 84 | x, 85 | bias_attr=None, 86 | # 指定偏置的属性的对象 87 | moving_mean_name=name + '3', 88 | # moving_mean的名称 89 | moving_variance_name=name + '4', 90 | # moving_variance的名称 91 | name=name, 92 | act=act, 93 | momentum=momentum, 94 | ) 95 | 96 | 97 | def resnetImagenet(input): 98 | #128 99 | x = layer_warp(basicblock, input, 64, 128, 1, 2) 100 | #64 101 | x = layer_warp(basicblock, x, 128, 256, 1, 2) 102 | #32 103 | x = layer_warp(basicblock, x, 256, 512, 1, 2) 104 | #16 105 | x = layer_warp(basicblock, x, 512, 1024, 1, 2) 106 | #8 107 | x = layer_warp(basicblock, x, 1024, 2048, 1, 2) 108 | #16 109 | x = deconv(x, num_filters=1024, filter_size=4, stride=2, padding=1) 110 | x = bn(x, name='bn_1', act='relu', momentum=0.5) 111 | #32 112 | x = deconv(x, num_filters=512, filter_size=4, stride=2, padding=1) 113 | x = bn(x, name='bn_2', act='relu', momentum=0.5) 114 | #64 115 | x = deconv(x, num_filters=256, filter_size=4, stride=2, padding=1) 116 | x = bn(x, name='bn_3', act='relu', momentum=0.5) 117 | #128 118 | x = deconv(x, num_filters=128, filter_size=4, stride=2, padding=1) 119 | x = bn(x, name='bn_4', act='relu', momentum=0.5) 120 | #256 121 | x = deconv(x, num_filters=64, filter_size=4, stride=2, padding=1) 122 | x = bn(x, name='bn_5', act='relu', momentum=0.5) 123 | 124 | x = deconv(x, num_filters=2, filter_size=3, stride=1, padding=1) 125 | return x 126 | 127 | 128 | def ResNettrain(): 129 | gray = fluid.layers.data(name='gray', shape=[1, 256, 256], dtype='float32') 130 | truth = fluid.layers.data(name='truth', shape=[2, 256, 256], dtype='float32') 131 | predict = resnetImagenet(gray) 132 | cost = createLoss(predict=predict, truth=truth) 133 | return predict, cost 134 | 135 | 136 | '''optimizer函数''' 137 | 138 | 139 | def optimizer_program(): 140 | return fluid.optimizer.Adam(learning_rate=2e-5, beta1=0.8) 141 | 142 | 143 | train_reader = paddle.batch(paddle.reader.shuffle( 144 | reader=train(), buf_size=7500 * 3 145 | ), batch_size=BATCH_SIZE) 146 | test_reader = paddle.batch(reader=test(), batch_size=10) 147 | 148 | use_cuda = True 149 | if not use_cuda: 150 | os.environ['CPU_NUM'] = str(6) 151 | feed_order = ['gray', 'weight'] 152 | place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace() 153 | 154 | main_program = fluid.default_main_program() 155 | star_program = fluid.default_startup_program() 156 | 157 | '''网络训练''' 158 | predict, cost = ResNettrain() 159 | 160 | '''优化函数''' 161 | optimizer = optimizer_program() 162 | optimizer.minimize(cost) 163 | 164 | exe = fluid.Executor(place) 165 | 166 | plt.ion() 167 | def train_loop(): 168 | gray = fluid.layers.data(name='gray', shape=[1, 256, 256], dtype='float32') 169 | truth = fluid.layers.data(name='truth', shape=[2, 256, 256], dtype='float32') 170 | feeder = fluid.DataFeeder( 171 | feed_list=['gray', 'truth'], place=place) 172 | exe.run(star_program) 173 | 174 | # 增量训练 175 | fluid.io.load_persistables(exe, 'work/model/incremental/', main_program) 176 | 177 | for pass_id in range(EPOCH_NUM): 178 | step = 0 179 | for data in train_reader(): 180 | loss = exe.run(main_program, feed=feeder.feed(data), fetch_list=[cost]) 181 | step += 1 182 | if step % 1000 == 0: 183 | try: 184 | generated_img = exe.run(main_program, feed=feeder.feed(data), fetch_list=[predict]) 185 | plt.figure(figsize=(15, 6)) 186 | plt.grid(False) 187 | for i in range(10): 188 | ab = generated_img[0][i] 189 | l = data[i][0][0] 190 | a = ab[0] 191 | b = ab[1] 192 | l = l[:, :, np.newaxis] 193 | a = a[:, :, np.newaxis].astype('float64') 194 | b = b[:, :, np.newaxis].astype('float64') 195 | lab = np.concatenate((l, a, b), axis=2) 196 | img = color.lab2rgb((lab)) 197 | img = transform.rotate(img, 270) 198 | img = np.fliplr(img) 199 | plt.grid(False) 200 | plt.subplot(2, 5, i + 1) 201 | plt.imshow(img) 202 | plt.axis('off') 203 | plt.xticks([]) 204 | plt.yticks([]) 205 | msg = 'Epoch ID={0} Batch ID={1} Loss={2}'.format(pass_id, step, loss[0][0]) 206 | plt.suptitle(msg, fontsize=20) 207 | plt.draw() 208 | plt.savefig('{}/{:04d}_{:04d}.png'.format('work/output_img', pass_id, step), bbox_inches='tight') 209 | plt.pause(0.01) 210 | except IOError: 211 | print(IOError) 212 | 213 | fluid.io.save_persistables(exe, 'work/model/incremental/', main_program) 214 | fluid.io.save_inference_model(Params_dirname, ["gray"], [predict], exe) 215 | plt.ioff() 216 | train_loop() -------------------------------------------------------------------------------- /train/Utils.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import matplotlib 3 | import math 4 | # matplotlib.use('agg') 5 | import matplotlib.pyplot as plt 6 | import matplotlib.gridspec as gridspec 7 | 8 | # % matplotlib inline 9 | # % config InlineBackend.figure_format = 'svg' 10 | 11 | img_dim = 48 12 | 13 | def plot(gen_data): 14 | pad_dim = 1 15 | paded = pad_dim + img_dim 16 | gen_data = gen_data.reshape(gen_data.shape[0], img_dim, img_dim) 17 | # math.ceil()函数返回数字的上入整数（取靠近0的整数） 18 | n = int(math.ceil(math.sqrt(gen_data.shape[0]))) 19 | # np.pad()方法：扩充矩阵，其中‘constant’指的是连续填充相同的值 20 | # np.transpose()函数用于对换数组的维度 21 | gen_data = (np.pad( 22 | gen_data, [[0, n * n - gen_data.shape[0]], [pad_dim, 0], [pad_dim, 0]], 23 | 'constant').reshape((n, n, paded, paded)).transpose((0, 2, 1, 3)) 24 | .reshape((n * paded, n * paded))) 25 | fig = plt.figure(figsize=(8, 8)) 26 | # plt.axis('off')中'off'表示关闭轴线和标签 27 | plt.axis('off') 28 | plt.imshow(gen_data, cmap='Greys_r', vmin=-1, vmax=1) 29 | return fig -------------------------------------------------------------------------------- /train/fileReader.py: -------------------------------------------------------------------------------- 1 | import os 2 | import cv2 3 | import numpy as np 4 | import paddle.dataset as dataset 5 | from skimage import io,color,transform 6 | import sklearn.neighbors as neighbors 7 | 8 | '''准备数据，定义Reader()''' 9 | 10 | PATH = 'work/train/' 11 | TEST = 'work/train/' 12 | 13 | 14 | class DataGenerater: 15 | def __init__(self): 16 | datalist = os.listdir(PATH) 17 | self.testlist = os.listdir(TEST) 18 | self.datalist = datalist 19 | 20 | def load(self, image): 21 | '''读取图片,并转为Lab，并提取出L和ab''' 22 | img = io.imread(image) 23 | lab = np.array(color.rgb2lab(img)).transpose() 24 | l = lab[:1, :, :] 25 | l = l.astype('float32') 26 | ab = lab[1:, :, :] 27 | ab = ab.astype('float32') 28 | return l, ab 29 | 30 | def create_train_reader(self): 31 | '''给dataset定义reader''' 32 | 33 | def reader(): 34 | for img in self.datalist: 35 | # print(img) 36 | try: 37 | l, ab = self.load(PATH + img) 38 | # print(ab) 39 | yield l.astype('float32'), ab.astype('float32') 40 | except Exception as e: 41 | print(e) 42 | 43 | return reader 44 | 45 | def create_test_reader(self, ): 46 | '''给test定义reader''' 47 | 48 | def reader(): 49 | for img in self.testlist: 50 | l, ab = self.load(TEST + img) 51 | yield l.astype('float32'), ab.astype('float32') 52 | 53 | return reader 54 | 55 | 56 | def train(batch_sizes=32): 57 | reader = DataGenerater().create_train_reader() 58 | return reader 59 | 60 | 61 | def test(): 62 | reader = DataGenerater().create_test_reader() 63 | return reader --------------------------------------------------------------------------------