├── output └── .keepgit.txt ├── example ├── PR_curve_each_class.png └── 20190808_041204.log.json_result.png ├── LICENSE ├── README.md ├── voc_eval_visualize.py ├── visualize.py ├── mean_ap_visualize.py └── json └── 20190808_041204.log.json /output/.keepgit.txt: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /example/PR_curve_each_class.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Stephenfang51/mmdetection_visualize/HEAD/example/PR_curve_each_class.png -------------------------------------------------------------------------------- /example/20190808_041204.log.json_result.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Stephenfang51/mmdetection_visualize/HEAD/example/20190808_041204.log.json_result.png -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2019 StephenFang 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # mmdetection_visualize_v1 2 | 3 | It's a very simple version for visualizing the training result produced by mmdetection 4 | 5 | ### Update 6 | > 2019.8.16 ----- PR_curve, F_measure for VOC dataset 7 | 8 | ### Readme 9 | The program supports drawing six training result and the most important evaluation tool:PR curve(only for VOC now) 10 | 11 | 1. loss_rpn_bbox 12 | 2. loss_rpn_cls 13 | 3. loss_bbox 14 | 4. loss_cls 15 | 5. loss 16 | 6. acc 17 | 18 | 19 | 7. PR_curve 20 | 8. F-measure 21 | 22 | 23 | ### Installation 24 | 25 | 1. Clone it 26 | `git clone https://github.com/Stephenfang51/mmdetection_visualize` 27 | 28 | There will be total 5 files(json directory, output directory, visualize.py, mean_ap_visualize.py, voc_eval_visualize.py) 29 | 30 | 2. 31 | - put `voc_eval_visualize.py` under `/mmdetection/tools/` 32 | 33 | - put `mean_ap_visualize.py` under `mmdetection/mmdet/core/evaluation/` 34 | 35 | ### How to use 36 | 37 | #### six training result 38 | 1. After training finished, you will have **work_dir** directory in your mmdetection directory 39 | 2. take the latest json file and put into json directory in mmditection_visualize directory 40 | 3. command `python visualize.py json/xxxxxxxlog.json` in terminal 41 | 4. check the output directory, Done ! 42 | 43 | #### PR curve and F-measure 44 | 1. make sure `voc_eval_visualize.py` and `mean_ap_visualize.py` settled down 45 | 2. command as usual like `python tools/voc_eval_visualize.py {your pkl file} {your network configs file}` 46 | - example `python tools/voc_eval_visualize.py result.pkl ./configs/faster_rcnn_r101_fpn_1x.py` 47 | 3. check the /mmdetection main directory, you will see the **PR_curve_each_class.png** there， Done ! 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | -------------------------------------------------------------------------------- /voc_eval_visualize.py: -------------------------------------------------------------------------------- 1 | from argparse import ArgumentParser 2 | 3 | import mmcv 4 | import numpy as np 5 | 6 | from mmdet import datasets 7 | from mmdet.core.evaluation.mean_ap_visualize import map_roc_pr 8 | 9 | 10 | def voc_eval(result_file, dataset, iou_thr=0.5): 11 | det_results = mmcv.load(result_file) 12 | gt_bboxes = [] 13 | gt_labels = [] 14 | gt_ignore = [] 15 | for i in range(len(dataset)): 16 | ann = dataset.get_ann_info(i) 17 | bboxes = ann['bboxes'] 18 | labels = ann['labels'] 19 | if 'bboxes_ignore' in ann: 20 | ignore = np.concatenate([ 21 | np.zeros(bboxes.shape[0], dtype=np.bool), 22 | np.ones(ann['bboxes_ignore'].shape[0], dtype=np.bool) 23 | ]) 24 | gt_ignore.append(ignore) 25 | bboxes = np.vstack([bboxes, ann['bboxes_ignore']]) 26 | labels = np.concatenate([labels, ann['labels_ignore']]) 27 | gt_bboxes.append(bboxes) 28 | gt_labels.append(labels) 29 | if not gt_ignore: 30 | gt_ignore = gt_ignore 31 | if hasattr(dataset, 'year') and dataset.year == 2007: 32 | dataset_name = 'voc07' 33 | else: 34 | dataset_name = dataset.CLASSES 35 | map_roc_pr( 36 | det_results, 37 | gt_bboxes, 38 | gt_labels, 39 | gt_ignore=gt_ignore, 40 | scale_ranges=None, 41 | iou_thr=iou_thr, 42 | dataset=dataset_name, 43 | print_summary=True) 44 | 45 | 46 | def main(): 47 | parser = ArgumentParser(description='VOC Evaluation') 48 | parser.add_argument('result', help='result file path') 49 | parser.add_argument('config', help='config file path') 50 | parser.add_argument( 51 | '--iou-thr', 52 | type=float, 53 | default=0.5, 54 | help='IoU threshold for evaluation') 55 | args = parser.parse_args() 56 | cfg = mmcv.Config.fromfile(args.config) 57 | test_dataset = mmcv.runner.obj_from_dict(cfg.data.test, datasets) 58 | voc_eval(args.result, test_dataset, args.iou_thr) 59 | 60 | 61 | if __name__ == '__main__': 62 | main() 63 | -------------------------------------------------------------------------------- /visualize.py: -------------------------------------------------------------------------------- 1 | import json 2 | import matplotlib.pyplot as plt 3 | import sys 4 | import os 5 | from collections import OrderedDict 6 | 7 | class visualize_mmdetection(): 8 | def __init__(self, path): 9 | self.log = open(path) 10 | self.dict_list = list() 11 | self.loss_rpn_bbox = list() 12 | self.loss_rpn_cls = list() 13 | self.loss_bbox = list() 14 | self.loss_cls = list() 15 | self.loss = list() 16 | self.acc = list() 17 | 18 | def load_data(self): 19 | for line in self.log: 20 | info = json.loads(line) 21 | self.dict_list.append(info) 22 | 23 | for i in range(1, len(self.dict_list)): 24 | for value, key in dict(self.dict_list[i]).items(): 25 | # ------------find key for every iter-------------------# 26 | loss_rpn_cls_value = dict(self.dict_list[i])['loss_rpn_cls'] 27 | loss_rpn_bbox_value = dict(self.dict_list[i])['loss_rpn_bbox'] 28 | loss_bbox_value = dict(self.dict_list[i])['loss_bbox'] 29 | loss_cls_value = dict(self.dict_list[i])['loss_cls'] 30 | loss_value = dict(self.dict_list[i])['loss'] 31 | acc_value = dict(self.dict_list[i])['acc'] 32 | # -------------list append------------------------------# 33 | self.loss_rpn_cls.append(loss_rpn_cls_value) 34 | self.loss_rpn_bbox.append(loss_rpn_bbox_value) 35 | self.loss_bbox.append(loss_bbox_value) 36 | self.loss_cls.append(loss_cls_value) 37 | self.loss.append(loss_value) 38 | self.acc.append(acc_value) 39 | # -------------clear repeated value---------------------# 40 | self.loss_rpn_cls = list(OrderedDict.fromkeys(self.loss_rpn_cls)) 41 | self.loss_rpn_bbox = list(OrderedDict.fromkeys(self.loss_rpn_bbox)) 42 | self.loss_bbox = list(OrderedDict.fromkeys(self.loss_bbox)) 43 | self.loss_cls = list(OrderedDict.fromkeys(self.loss_cls)) 44 | self.loss = list(OrderedDict.fromkeys(self.loss)) 45 | self.acc = list(OrderedDict.fromkeys(self.acc)) 46 | 47 | def show_chart(self): 48 | plt.rcParams.update({'font.size': 15}) 49 | 50 | plt.figure(figsize=(20, 20)) 51 | 52 | plt.subplot(321, title='loss_rpn_cls', ylabel='loss') 53 | plt.plot(self.loss_rpn_cls) 54 | plt.subplot(322, title='loss_rpn_bbox', ylabel='loss') 55 | plt.plot(self.loss_rpn_bbox) 56 | 57 | plt.subplot(323, title='loss_cls', ylabel='loss') 58 | plt.plot(self.loss_cls) 59 | plt.subplot(324, title='loss_bbox', ylabel='loss') 60 | plt.plot(self.loss_bbox) 61 | plt.subplot(325, title='total loss', ylabel='loss') 62 | plt.plot(self.loss) 63 | plt.subplot(326, title='accuracy', ylabel='accuracy') 64 | plt.plot(self.acc) 65 | plt.suptitle((sys.argv[1][5:] + "\n training result"), fontsize=30) 66 | plt.savefig(('output/' + sys.argv[1][5:] + '_result.png')) 67 | 68 | 69 | if __name__ == '__main__': 70 | x = visualize_mmdetection(sys.argv[1]) 71 | x.load_data() 72 | x.show_chart() -------------------------------------------------------------------------------- /mean_ap_visualize.py: -------------------------------------------------------------------------------- 1 | import mmcv 2 | import numpy as np 3 | import matplotlib.pyplot as plt 4 | from terminaltables import AsciiTable 5 | 6 | from .bbox_overlaps import bbox_overlaps 7 | from .class_names import get_classes 8 | 9 | 10 | def average_precision(recalls, precisions, mode='area'): 11 | """Calculate average precision (for single or multiple scales). 12 | 13 | Args: 14 | recalls (ndarray): shape (num_scales, num_dets) or (num_dets, ) 15 | precisions (ndarray): shape (num_scales, num_dets) or (num_dets, ) 16 | mode (str): 'area' or '11points', 'area' means calculating the area 17 | under precision-recall curve, '11points' means calculating 18 | the average precision of recalls at [0, 0.1, ..., 1] 19 | 20 | Returns: 21 | float or ndarray: calculated average precision 22 | """ 23 | no_scale = False 24 | if recalls.ndim == 1: 25 | no_scale = True 26 | recalls = recalls[np.newaxis, :] 27 | precisions = precisions[np.newaxis, :] 28 | assert recalls.shape == precisions.shape and recalls.ndim == 2 29 | num_scales = recalls.shape[0] 30 | ap = np.zeros(num_scales, dtype=np.float32) 31 | if mode == 'area': 32 | zeros = np.zeros((num_scales, 1), dtype=recalls.dtype) 33 | ones = np.ones((num_scales, 1), dtype=recalls.dtype) 34 | mrec = np.hstack((zeros, recalls, ones)) 35 | mpre = np.hstack((zeros, precisions, zeros)) 36 | for i in range(mpre.shape[1] - 1, 0, -1): 37 | mpre[:, i - 1] = np.maximum(mpre[:, i - 1], mpre[:, i]) 38 | for i in range(num_scales): 39 | ind = np.where(mrec[i, 1:] != mrec[i, :-1])[0] 40 | ap[i] = np.sum( 41 | (mrec[i, ind + 1] - mrec[i, ind]) * mpre[i, ind + 1]) 42 | elif mode == '11points': 43 | for i in range(num_scales): 44 | for thr in np.arange(0, 1 + 1e-3, 0.1): 45 | precs = precisions[i, recalls[i, :] >= thr] 46 | prec = precs.max() if precs.size > 0 else 0 47 | ap[i] += prec 48 | ap /= 11 49 | else: 50 | raise ValueError( 51 | 'Unrecognized mode, only "area" and "11points" are supported') 52 | if no_scale: 53 | ap = ap[0] 54 | return ap 55 | 56 | 57 | def tpfp_imagenet(det_bboxes, 58 | gt_bboxes, 59 | gt_ignore, 60 | default_iou_thr, 61 | area_ranges=None): 62 | """Check if detected bboxes are true positive or false positive. 63 | 64 | Args: 65 | det_bbox (ndarray): the detected bbox 66 | gt_bboxes (ndarray): ground truth bboxes of this image 67 | gt_ignore (ndarray): indicate if gts are ignored for evaluation or not 68 | default_iou_thr (float): the iou thresholds for medium and large bboxes 69 | area_ranges (list or None): gt bbox area ranges 70 | 71 | Returns: 72 | tuple: two arrays (tp, fp) whose elements are 0 and 1 73 | """ 74 | num_dets = det_bboxes.shape[0] 75 | num_gts = gt_bboxes.shape[0] 76 | if area_ranges is None: 77 | area_ranges = [(None, None)] 78 | num_scales = len(area_ranges) 79 | # tp and fp are of shape (num_scales, num_gts), each row is tp or fp 80 | # of a certain scale. 81 | tp = np.zeros((num_scales, num_dets), dtype=np.float32) 82 | fp = np.zeros((num_scales, num_dets), dtype=np.float32) 83 | if gt_bboxes.shape[0] == 0: 84 | if area_ranges == [(None, None)]: 85 | fp[...] = 1 86 | else: 87 | det_areas = (det_bboxes[:, 2] - det_bboxes[:, 0] + 1) * ( 88 | det_bboxes[:, 3] - det_bboxes[:, 1] + 1) 89 | for i, (min_area, max_area) in enumerate(area_ranges): 90 | fp[i, (det_areas >= min_area) & (det_areas < max_area)] = 1 91 | return tp, fp 92 | ious = bbox_overlaps(det_bboxes, gt_bboxes - 1) 93 | gt_w = gt_bboxes[:, 2] - gt_bboxes[:, 0] + 1 94 | gt_h = gt_bboxes[:, 3] - gt_bboxes[:, 1] + 1 95 | iou_thrs = np.minimum((gt_w * gt_h) / ((gt_w + 10.0) * (gt_h + 10.0)), 96 | default_iou_thr) 97 | # sort all detections by scores in descending order 98 | sort_inds = np.argsort(-det_bboxes[:, -1]) 99 | for k, (min_area, max_area) in enumerate(area_ranges): 100 | gt_covered = np.zeros(num_gts, dtype=bool) 101 | # if no area range is specified, gt_area_ignore is all False 102 | if min_area is None: 103 | gt_area_ignore = np.zeros_like(gt_ignore, dtype=bool) 104 | else: 105 | gt_areas = gt_w * gt_h 106 | gt_area_ignore = (gt_areas < min_area) | (gt_areas >= max_area) 107 | for i in sort_inds: 108 | max_iou = -1 109 | matched_gt = -1 110 | # find best overlapped available gt 111 | for j in range(num_gts): 112 | # different from PASCAL VOC: allow finding other gts if the 113 | # best overlaped ones are already matched by other det bboxes 114 | if gt_covered[j]: 115 | continue 116 | elif ious[i, j] >= iou_thrs[j] and ious[i, j] > max_iou: 117 | max_iou = ious[i, j] 118 | matched_gt = j 119 | # there are 4 cases for a det bbox: 120 | # 1. it matches a gt, tp = 1, fp = 0 121 | # 2. it matches an ignored gt, tp = 0, fp = 0 122 | # 3. it matches no gt and within area range, tp = 0, fp = 1 123 | # 4. it matches no gt but is beyond area range, tp = 0, fp = 0 124 | if matched_gt >= 0: 125 | gt_covered[matched_gt] = 1 126 | if not (gt_ignore[matched_gt] or gt_area_ignore[matched_gt]): 127 | tp[k, i] = 1 128 | elif min_area is None: 129 | fp[k, i] = 1 130 | else: 131 | bbox = det_bboxes[i, :4] 132 | area = (bbox[2] - bbox[0] + 1) * (bbox[3] - bbox[1] + 1) 133 | if area >= min_area and area < max_area: 134 | fp[k, i] = 1 135 | return tp, fp 136 | 137 | 138 | def tpfp_default(det_bboxes, gt_bboxes, gt_ignore, iou_thr, area_ranges=None): 139 | """Check if detected bboxes are true positive or false positive. 140 | 141 | Args: 142 | det_bbox (ndarray): the detected bbox 143 | gt_bboxes (ndarray): ground truth bboxes of this image 144 | gt_ignore (ndarray): indicate if gts are ignored for evaluation or not 145 | iou_thr (float): the iou thresholds 146 | 147 | Returns: 148 | tuple: (tp, fp), two arrays whose elements are 0 and 1 149 | """ 150 | num_dets = det_bboxes.shape[0] 151 | num_gts = gt_bboxes.shape[0] 152 | if area_ranges is None: 153 | area_ranges = [(None, None)] 154 | num_scales = len(area_ranges) 155 | # tp and fp are of shape (num_scales, num_gts), each row is tp or fp of 156 | # a certain scale 157 | tp = np.zeros((num_scales, num_dets), dtype=np.float32) 158 | fp = np.zeros((num_scales, num_dets), dtype=np.float32) 159 | # if there is no gt bboxes in this image, then all det bboxes 160 | # within area range are false positives 161 | if gt_bboxes.shape[0] == 0: 162 | if area_ranges == [(None, None)]: 163 | fp[...] = 1 164 | else: 165 | det_areas = (det_bboxes[:, 2] - det_bboxes[:, 0] + 1) * ( 166 | det_bboxes[:, 3] - det_bboxes[:, 1] + 1) 167 | for i, (min_area, max_area) in enumerate(area_ranges): 168 | fp[i, (det_areas >= min_area) & (det_areas < max_area)] = 1 169 | return tp, fp 170 | ious = bbox_overlaps(det_bboxes, gt_bboxes) 171 | ious_max = ious.max(axis=1) 172 | ious_argmax = ious.argmax(axis=1) 173 | sort_inds = np.argsort(-det_bboxes[:, -1]) 174 | for k, (min_area, max_area) in enumerate(area_ranges): 175 | gt_covered = np.zeros(num_gts, dtype=bool) 176 | # if no area range is specified, gt_area_ignore is all False 177 | if min_area is None: 178 | gt_area_ignore = np.zeros_like(gt_ignore, dtype=bool) 179 | else: 180 | gt_areas = (gt_bboxes[:, 2] - gt_bboxes[:, 0] + 1) * ( 181 | gt_bboxes[:, 3] - gt_bboxes[:, 1] + 1) 182 | gt_area_ignore = (gt_areas < min_area) | (gt_areas >= max_area) 183 | for i in sort_inds: 184 | if ious_max[i] >= iou_thr: 185 | matched_gt = ious_argmax[i] 186 | if not (gt_ignore[matched_gt] or gt_area_ignore[matched_gt]): 187 | if not gt_covered[matched_gt]: 188 | gt_covered[matched_gt] = True 189 | tp[k, i] = 1 190 | else: 191 | fp[k, i] = 1 192 | # otherwise ignore this detected bbox, tp = 0, fp = 0 193 | elif min_area is None: 194 | fp[k, i] = 1 195 | else: 196 | bbox = det_bboxes[i, :4] 197 | area = (bbox[2] - bbox[0] + 1) * (bbox[3] - bbox[1] + 1) 198 | if area >= min_area and area < max_area: 199 | fp[k, i] = 1 200 | return tp, fp 201 | 202 | 203 | def get_cls_results(det_results, gt_bboxes, gt_labels, gt_ignore, class_id): 204 | """Get det results and gt information of a certain class.""" 205 | cls_dets = [det[class_id] 206 | for det in det_results] # det bboxes of this class 207 | cls_gts = [] # gt bboxes of this class 208 | cls_gt_ignore = [] 209 | for j in range(len(gt_bboxes)): 210 | gt_bbox = gt_bboxes[j] 211 | cls_inds = (gt_labels[j] == class_id + 1) 212 | cls_gt = gt_bbox[cls_inds, :] if gt_bbox.shape[0] > 0 else gt_bbox 213 | cls_gts.append(cls_gt) 214 | if gt_ignore is None: 215 | cls_gt_ignore.append(np.zeros(cls_gt.shape[0], dtype=np.int32)) 216 | else: 217 | cls_gt_ignore.append(gt_ignore[j][cls_inds]) 218 | return cls_dets, cls_gts, cls_gt_ignore 219 | 220 | 221 | def map_roc_pr(det_results, 222 | gt_bboxes, 223 | gt_labels, 224 | gt_ignore=None, 225 | scale_ranges=None, 226 | iou_thr=0.5, 227 | dataset=None, 228 | print_summary=True): 229 | """Evaluate mAP of a dataset. 230 | 231 | Args: 232 | det_results (list): a list of list, [[cls1_det, cls2_det, ...], ...] 233 | gt_bboxes (list): ground truth bboxes of each image, a list of K*4 234 | array. 235 | gt_labels (list): ground truth labels of each image, a list of K array 236 | gt_ignore (list): gt ignore indicators of each image, a list of K array 237 | scale_ranges (list, optional): [(min1, max1), (min2, max2), ...] 238 | iou_thr (float): IoU threshold 239 | dataset (None or str or list): dataset name or dataset classes, there 240 | are minor differences in metrics for different datsets, e.g. 241 | "voc07", "imagenet_det", etc. 242 | print_summary (bool): whether to print the mAP summary 243 | 244 | Returns: 245 | tuple: (mAP, [dict, dict, ...]) 246 | """ 247 | assert len(det_results) == len(gt_bboxes) == len(gt_labels) 248 | if gt_ignore is not None: 249 | assert len(gt_ignore) == len(gt_labels) 250 | for i in range(len(gt_ignore)): 251 | assert len(gt_labels[i]) == len(gt_ignore[i]) 252 | area_ranges = ([(rg[0]**2, rg[1]**2) for rg in scale_ranges] 253 | if scale_ranges is not None else None) 254 | num_scales = len(scale_ranges) if scale_ranges is not None else 1 255 | eval_results = [] 256 | num_classes = len(det_results[0]) # positive class num 257 | gt_labels = [ 258 | label if label.ndim == 1 else label[:, 0] for label in gt_labels 259 | ] 260 | 261 | f_measure_list = [] 262 | recall_list = [] 263 | precision_list = [] 264 | ap_list = [] 265 | for i in range(num_classes): 266 | # get gt and det bboxes of this class 267 | cls_dets, cls_gts, cls_gt_ignore = get_cls_results( 268 | det_results, gt_bboxes, gt_labels, gt_ignore, i) 269 | # calculate tp and fp for each image 270 | tpfp_func = ( 271 | tpfp_imagenet if dataset in ['det', 'vid'] else tpfp_default) 272 | tpfp = [ 273 | tpfp_func(cls_dets[j], cls_gts[j], cls_gt_ignore[j], iou_thr, 274 | area_ranges) for j in range(len(cls_dets)) 275 | ] 276 | tp, fp = tuple(zip(*tpfp)) 277 | # calculate gt number of each scale, gts ignored or beyond scale 278 | # are not counted 279 | num_gts = np.zeros(num_scales, dtype=int) 280 | for j, bbox in enumerate(cls_gts): 281 | if area_ranges is None: 282 | num_gts[0] += np.sum(np.logical_not(cls_gt_ignore[j])) 283 | else: 284 | gt_areas = (bbox[:, 2] - bbox[:, 0] + 1) * ( 285 | bbox[:, 3] - bbox[:, 1] + 1) 286 | for k, (min_area, max_area) in enumerate(area_ranges): 287 | num_gts[k] += np.sum( 288 | np.logical_not(cls_gt_ignore[j]) & 289 | (gt_areas >= min_area) & (gt_areas < max_area)) 290 | # sort all det bboxes by score, also sort tp and fp 291 | cls_dets = np.vstack(cls_dets) 292 | num_dets = cls_dets.shape[0] 293 | sort_inds = np.argsort(-cls_dets[:, -1]) 294 | tp = np.hstack(tp)[:, sort_inds] 295 | fp = np.hstack(fp)[:, sort_inds] 296 | # calculate recall and precision with tp and fp 297 | tp = np.cumsum(tp, axis=1) 298 | fp = np.cumsum(fp, axis=1) 299 | eps = np.finfo(np.float32).eps 300 | recalls = tp / np.maximum(num_gts[:, np.newaxis], eps) 301 | precisions = tp / np.maximum((tp + fp), eps) 302 | # calculate AP 303 | 304 | 305 | if scale_ranges is None: 306 | recalls = recalls[0, :] 307 | precisions = precisions[0, :] 308 | num_gts = num_gts.item() 309 | mode = 'area' if dataset != 'voc07' else '11points' 310 | 311 | ap = average_precision(recalls, precisions, mode) 312 | 313 | 314 | #------------collect data--------------# 315 | recall_list.append(recalls) 316 | precision_list.append(precisions) 317 | ap_list.append(ap) 318 | 319 | 320 | 321 | eval_results.append({ 322 | 'num_gts': num_gts, 323 | 'num_dets': num_dets, 324 | 'recall': recalls, 325 | 'precision': precisions, 326 | 'ap': ap 327 | }) 328 | 329 | #-----------F_measure calculation--------------# 330 | 331 | top = recalls * precisions 332 | down = recalls + precisions 333 | f_measure = np.mean(2*(top/down)) 334 | f_measure_list.append(f_measure) 335 | 336 | label_names = get_classes(dataset) 337 | #------------plot PR / F-measure-----------# 338 | plt.figure(figsize=(6, 10)) 339 | #------------plot curve--------------------# 340 | for i in range(num_classes): 341 | plt.subplots_adjust(hspace=0.3) 342 | plt.subplot(211) 343 | plt.plot(recall_list[i], precision_list[i], linewidth=2, \ 344 | label='({}, (AP = {:.4f}))'.format(label_names[i], ap_list[i])) 345 | plt.title('Precision-Recall') 346 | plt.xlabel('Recall') 347 | plt.ylabel('Precision') 348 | plt.axis([0, 1, 0, 1]) 349 | plt.legend() 350 | 351 | plt.subplot(212) 352 | plt.title('F-measure') 353 | plt.bar(label_names[i], f_measure_list[i]) 354 | for a, b in zip(label_names, f_measure_list): 355 | plt.text(a, b, '%.4f' % b, color='black', fontweight='bold') 356 | 357 | plt.savefig('/mmdetection/PR_Curve_each_class.png') 358 | 359 | if scale_ranges is not None: 360 | # shape (num_classes, num_scales) 361 | all_ap = np.vstack([cls_result['ap'] for cls_result in eval_results]) 362 | all_num_gts = np.vstack( 363 | [cls_result['num_gts'] for cls_result in eval_results]) 364 | mean_ap = [ 365 | all_ap[all_num_gts[:, i] > 0, i].mean() 366 | if np.any(all_num_gts[:, i] > 0) else 0.0 367 | for i in range(num_scales) 368 | ] 369 | else: 370 | aps = [] 371 | for cls_result in eval_results: 372 | if cls_result['num_gts'] > 0: 373 | aps.append(cls_result['ap']) 374 | mean_ap = np.array(aps).mean().item() if aps else 0.0 375 | if print_summary: 376 | pass 377 | # print_map_summary(mean_ap, eval_results, dataset) 378 | 379 | return mean_ap, eval_results 380 | 381 | 382 | def print_map_summary(mean_ap, results, dataset=None): 383 | """Print mAP and results of each class. 384 | 385 | Args: 386 | mean_ap(float): calculated from `eval_map` 387 | results(list): calculated from `eval_map` 388 | dataset(None or str or list): dataset name or dataset classes. 389 | """ 390 | num_scales = len(results[0]['ap']) if isinstance(results[0]['ap'], 391 | np.ndarray) else 1 392 | num_classes = len(results) 393 | 394 | recalls = np.zeros((num_scales, num_classes), dtype=np.float32) 395 | precisions = np.zeros((num_scales, num_classes), dtype=np.float32) 396 | aps = np.zeros((num_scales, num_classes), dtype=np.float32) 397 | num_gts = np.zeros((num_scales, num_classes), dtype=int) 398 | for i, cls_result in enumerate(results): 399 | if cls_result['recall'].size > 0: 400 | recalls[:, i] = np.array(cls_result['recall'], ndmin=2)[:, -1] 401 | precisions[:, i] = np.array( 402 | cls_result['precision'], ndmin=2)[:, -1] 403 | aps[:, i] = cls_result['ap'] 404 | num_gts[:, i] = cls_result['num_gts'] 405 | 406 | if dataset is None: 407 | label_names = [str(i) for i in range(1, num_classes + 1)] 408 | elif mmcv.is_str(dataset): 409 | label_names = get_classes(dataset) 410 | else: 411 | label_names = dataset 412 | 413 | 414 | 415 | if not isinstance(mean_ap, list): 416 | mean_ap = [mean_ap] 417 | header = ['class', 'gts', 'dets', 'recall', 'precision', 'ap'] 418 | for i in range(num_scales): 419 | table_data = [header] 420 | for j in range(num_classes): 421 | row_data = [ 422 | label_names[j], num_gts[i, j], results[j]['num_dets'], 423 | '{:.3f}'.format(recalls[i, j]), '{:.3f}'.format( 424 | precisions[i, j]), '{:.3f}'.format(aps[i, j]) 425 | ] 426 | table_data.append(row_data) 427 | table_data.append(['mAP', '', '', '', '', '{:.3f}'.format(mean_ap[i])]) 428 | table = AsciiTable(table_data) 429 | table.inner_footing_row_border = True 430 | print(table.table) 431 | -------------------------------------------------------------------------------- /json/20190808_041204.log.json: -------------------------------------------------------------------------------- 1 | {"mode": "train", "epoch": 1, "iter": 50, "lr": 0.00797, "time": 0.59679, "data_time": 0.0968, "memory": 5693, "loss_rpn_cls": 0.11835, "loss_rpn_bbox": 0.01221, "loss_cls": 0.49759, "acc": 94.61719, "loss_bbox": 0.08404, "loss": 0.71218} 2 | {"mode": "train", "epoch": 1, "iter": 100, "lr": 0.00931, "time": 0.57119, "data_time": 0.0656, "memory": 5693, "loss_rpn_cls": 0.05051, "loss_rpn_bbox": 0.00928, "loss_cls": 0.196, "acc": 96.15625, "loss_bbox": 0.09498, "loss": 0.35077} 3 | {"mode": "train", "epoch": 1, "iter": 150, "lr": 0.01064, "time": 0.59005, "data_time": 0.00892, "memory": 5693, "loss_rpn_cls": 0.03875, "loss_rpn_bbox": 0.00888, "loss_cls": 0.20359, "acc": 95.40625, "loss_bbox": 0.10541, "loss": 0.35663} 4 | {"mode": "train", "epoch": 1, "iter": 200, "lr": 0.01197, "time": 0.59531, "data_time": 0.08774, "memory": 5693, "loss_rpn_cls": 0.03968, "loss_rpn_bbox": 0.01054, "loss_cls": 0.21803, "acc": 95.37305, "loss_bbox": 0.12022, "loss": 0.38847} 5 | {"mode": "train", "epoch": 1, "iter": 250, "lr": 0.01331, "time": 0.61337, "data_time": 0.07735, "memory": 5693, "loss_rpn_cls": 0.03748, "loss_rpn_bbox": 0.00867, "loss_cls": 0.21153, "acc": 95.86914, "loss_bbox": 0.10046, "loss": 0.35815} 6 | {"mode": "train", "epoch": 1, "iter": 300, "lr": 0.01464, "time": 0.79358, "data_time": 0.24043, "memory": 5693, "loss_rpn_cls": 0.0348, "loss_rpn_bbox": 0.00874, "loss_cls": 0.24501, "acc": 95.40234, "loss_bbox": 0.12003, "loss": 0.40859} 7 | {"mode": "train", "epoch": 1, "iter": 350, "lr": 0.01597, "time": 0.54154, "data_time": 0.03431, "memory": 5693, "loss_rpn_cls": 0.02722, "loss_rpn_bbox": 0.00876, "loss_cls": 0.1903, "acc": 95.93555, "loss_bbox": 0.10312, "loss": 0.3294} 8 | {"mode": "train", "epoch": 1, "iter": 400, "lr": 0.01731, "time": 0.60817, "data_time": 0.04485, "memory": 5693, "loss_rpn_cls": 0.02287, "loss_rpn_bbox": 0.00667, "loss_cls": 0.16813, "acc": 96.03516, "loss_bbox": 0.1006, "loss": 0.29827} 9 | {"mode": "train", "epoch": 1, "iter": 450, "lr": 0.01864, "time": 0.52028, "data_time": 0.01367, "memory": 5696, "loss_rpn_cls": 0.03169, "loss_rpn_bbox": 0.00844, "loss_cls": 0.20744, "acc": 95.05664, "loss_bbox": 0.11935, "loss": 0.36693} 10 | {"mode": "train", "epoch": 1, "iter": 500, "lr": 0.01997, "time": 0.51426, "data_time": 0.00634, "memory": 5696, "loss_rpn_cls": 0.0309, "loss_rpn_bbox": 0.00783, "loss_cls": 0.17481, "acc": 96.0, "loss_bbox": 0.10127, "loss": 0.31481} 11 | {"mode": "train", "epoch": 1, "iter": 550, "lr": 0.02, "time": 0.63824, "data_time": 0.08104, "memory": 5696, "loss_rpn_cls": 0.03039, "loss_rpn_bbox": 0.00915, "loss_cls": 0.22103, "acc": 95.04492, "loss_bbox": 0.11644, "loss": 0.377} 12 | {"mode": "train", "epoch": 2, "iter": 50, "lr": 0.02, "time": 0.71032, "data_time": 0.08964, "memory": 5696, "loss_rpn_cls": 0.03301, "loss_rpn_bbox": 0.01058, "loss_cls": 0.17785, "acc": 95.66406, "loss_bbox": 0.10286, "loss": 0.3243} 13 | {"mode": "train", "epoch": 2, "iter": 100, "lr": 0.02, "time": 0.51818, "data_time": 0.00659, "memory": 5696, "loss_rpn_cls": 0.0208, "loss_rpn_bbox": 0.00904, "loss_cls": 0.19729, "acc": 95.32617, "loss_bbox": 0.11022, "loss": 0.33734} 14 | {"mode": "train", "epoch": 2, "iter": 150, "lr": 0.02, "time": 0.51592, "data_time": 0.00659, "memory": 5696, "loss_rpn_cls": 0.03901, "loss_rpn_bbox": 0.01021, "loss_cls": 0.23754, "acc": 94.82812, "loss_bbox": 0.11851, "loss": 0.40527} 15 | {"mode": "train", "epoch": 2, "iter": 200, "lr": 0.02, "time": 0.73885, "data_time": 0.14225, "memory": 5696, "loss_rpn_cls": 0.02451, "loss_rpn_bbox": 0.00861, "loss_cls": 0.19385, "acc": 95.19727, "loss_bbox": 0.10589, "loss": 0.33286} 16 | {"mode": "train", "epoch": 2, "iter": 250, "lr": 0.02, "time": 0.67018, "data_time": 0.05752, "memory": 5696, "loss_rpn_cls": 0.02127, "loss_rpn_bbox": 0.00676, "loss_cls": 0.21145, "acc": 95.10742, "loss_bbox": 0.11188, "loss": 0.35136} 17 | {"mode": "train", "epoch": 2, "iter": 300, "lr": 0.02, "time": 1.77583, "data_time": 1.21891, "memory": 5696, "loss_rpn_cls": 0.02736, "loss_rpn_bbox": 0.00849, "loss_cls": 0.22782, "acc": 95.00586, "loss_bbox": 0.11242, "loss": 0.37608} 18 | {"mode": "train", "epoch": 2, "iter": 350, "lr": 0.02, "time": 0.58246, "data_time": 0.01317, "memory": 5696, "loss_rpn_cls": 0.02791, "loss_rpn_bbox": 0.00913, "loss_cls": 0.15841, "acc": 95.70312, "loss_bbox": 0.09111, "loss": 0.28656} 19 | {"mode": "train", "epoch": 2, "iter": 400, "lr": 0.02, "time": 1.81113, "data_time": 1.26989, "memory": 5696, "loss_rpn_cls": 0.02869, "loss_rpn_bbox": 0.00838, "loss_cls": 0.22194, "acc": 94.375, "loss_bbox": 0.11855, "loss": 0.37758} 20 | {"mode": "train", "epoch": 2, "iter": 450, "lr": 0.02, "time": 0.55255, "data_time": 0.00809, "memory": 5696, "loss_rpn_cls": 0.02299, "loss_rpn_bbox": 0.00882, "loss_cls": 0.19402, "acc": 95.03125, "loss_bbox": 0.10782, "loss": 0.33365} 21 | {"mode": "train", "epoch": 2, "iter": 500, "lr": 0.02, "time": 0.52904, "data_time": 0.02711, "memory": 5696, "loss_rpn_cls": 0.02585, "loss_rpn_bbox": 0.00886, "loss_cls": 0.15981, "acc": 95.43945, "loss_bbox": 0.08673, "loss": 0.28125} 22 | {"mode": "train", "epoch": 2, "iter": 550, "lr": 0.02, "time": 0.50466, "data_time": 0.0066, "memory": 5696, "loss_rpn_cls": 0.01482, "loss_rpn_bbox": 0.00775, "loss_cls": 0.16066, "acc": 95.36328, "loss_bbox": 0.09243, "loss": 0.27565} 23 | {"mode": "train", "epoch": 3, "iter": 115, "lr": 0.02, "time": 0.4643, "data_time": 0.18293, "memory": 5696, "loss_rpn_cls": 0.015, "loss_rpn_bbox": 0.00872, "loss_cls": 0.16652, "acc": 95.33798, "loss_bbox": 0.08805, "loss": 0.2783} 24 | {"mode": "train", "epoch": 3, "iter": 50, "lr": 0.02, "time": 0.65571, "data_time": 0.042, "memory": 5696, "loss_rpn_cls": 0.01512, "loss_rpn_bbox": 0.00929, "loss_cls": 0.19279, "acc": 94.68555, "loss_bbox": 0.10084, "loss": 0.31803} 25 | {"mode": "train", "epoch": 3, "iter": 100, "lr": 0.02, "time": 0.62714, "data_time": 0.00941, "memory": 5696, "loss_rpn_cls": 0.01642, "loss_rpn_bbox": 0.00869, "loss_cls": 0.18565, "acc": 94.45898, "loss_bbox": 0.10781, "loss": 0.31857} 26 | {"mode": "train", "epoch": 3, "iter": 150, "lr": 0.02, "time": 0.53399, "data_time": 0.00724, "memory": 5696, "loss_rpn_cls": 0.0136, "loss_rpn_bbox": 0.00865, "loss_cls": 0.16028, "acc": 94.70508, "loss_bbox": 0.0894, "loss": 0.27193} 27 | {"mode": "train", "epoch": 3, "iter": 200, "lr": 0.02, "time": 0.51833, "data_time": 0.01012, "memory": 5696, "loss_rpn_cls": 0.01231, "loss_rpn_bbox": 0.00654, "loss_cls": 0.15808, "acc": 95.64258, "loss_bbox": 0.07831, "loss": 0.25524} 28 | {"mode": "train", "epoch": 3, "iter": 250, "lr": 0.02, "time": 0.51886, "data_time": 0.01523, "memory": 5696, "loss_rpn_cls": 0.01818, "loss_rpn_bbox": 0.00802, "loss_cls": 0.16348, "acc": 95.30078, "loss_bbox": 0.08176, "loss": 0.27143} 29 | {"mode": "train", "epoch": 3, "iter": 300, "lr": 0.02, "time": 0.52, "data_time": 0.01675, "memory": 5696, "loss_rpn_cls": 0.01862, "loss_rpn_bbox": 0.01024, "loss_cls": 0.19063, "acc": 94.18359, "loss_bbox": 0.10709, "loss": 0.32658} 30 | {"mode": "train", "epoch": 3, "iter": 350, "lr": 0.02, "time": 0.51337, "data_time": 0.01368, "memory": 5696, "loss_rpn_cls": 0.01528, "loss_rpn_bbox": 0.00834, "loss_cls": 0.17178, "acc": 94.70508, "loss_bbox": 0.09347, "loss": 0.28888} 31 | {"mode": "train", "epoch": 3, "iter": 400, "lr": 0.02, "time": 0.52096, "data_time": 0.01297, "memory": 5696, "loss_rpn_cls": 0.01524, "loss_rpn_bbox": 0.00654, "loss_cls": 0.13255, "acc": 95.80273, "loss_bbox": 0.07087, "loss": 0.22519} 32 | {"mode": "train", "epoch": 3, "iter": 450, "lr": 0.02, "time": 0.52238, "data_time": 0.01699, "memory": 5696, "loss_rpn_cls": 0.01728, "loss_rpn_bbox": 0.00808, "loss_cls": 0.1352, "acc": 96.00391, "loss_bbox": 0.07413, "loss": 0.23469} 33 | {"mode": "train", "epoch": 3, "iter": 500, "lr": 0.02, "time": 0.54049, "data_time": 0.03636, "memory": 5696, "loss_rpn_cls": 0.02972, "loss_rpn_bbox": 0.0097, "loss_cls": 0.17275, "acc": 95.46094, "loss_bbox": 0.08518, "loss": 0.29735} 34 | {"mode": "train", "epoch": 3, "iter": 550, "lr": 0.02, "time": 0.81261, "data_time": 0.31663, "memory": 5696, "loss_rpn_cls": 0.0269, "loss_rpn_bbox": 0.00791, "loss_cls": 0.15391, "acc": 95.45117, "loss_bbox": 0.08503, "loss": 0.27375} 35 | {"mode": "train", "epoch": 4, "iter": 50, "lr": 0.02, "time": 0.60127, "data_time": 0.09988, "memory": 5696, "loss_rpn_cls": 0.01311, "loss_rpn_bbox": 0.00816, "loss_cls": 0.14259, "acc": 95.31836, "loss_bbox": 0.07658, "loss": 0.24045} 36 | {"mode": "train", "epoch": 4, "iter": 100, "lr": 0.02, "time": 0.71163, "data_time": 0.17248, "memory": 5696, "loss_rpn_cls": 0.01104, "loss_rpn_bbox": 0.00651, "loss_cls": 0.15079, "acc": 95.00586, "loss_bbox": 0.07959, "loss": 0.24793} 37 | {"mode": "train", "epoch": 4, "iter": 150, "lr": 0.02, "time": 0.57527, "data_time": 0.03136, "memory": 5696, "loss_rpn_cls": 0.01315, "loss_rpn_bbox": 0.00905, "loss_cls": 0.15083, "acc": 95.10742, "loss_bbox": 0.07495, "loss": 0.24799} 38 | {"mode": "train", "epoch": 4, "iter": 200, "lr": 0.02, "time": 0.56739, "data_time": 0.03235, "memory": 5696, "loss_rpn_cls": 0.01052, "loss_rpn_bbox": 0.00567, "loss_cls": 0.11691, "acc": 95.96094, "loss_bbox": 0.05841, "loss": 0.19151} 39 | {"mode": "train", "epoch": 4, "iter": 250, "lr": 0.02, "time": 0.6406, "data_time": 0.11829, "memory": 5696, "loss_rpn_cls": 0.03531, "loss_rpn_bbox": 0.01054, "loss_cls": 0.25537, "acc": 94.49609, "loss_bbox": 0.107, "loss": 0.40822} 40 | {"mode": "train", "epoch": 4, "iter": 300, "lr": 0.02, "time": 0.51271, "data_time": 0.00934, "memory": 5696, "loss_rpn_cls": 0.03043, "loss_rpn_bbox": 0.00864, "loss_cls": 0.20869, "acc": 95.01367, "loss_bbox": 0.10187, "loss": 0.34963} 41 | {"mode": "train", "epoch": 4, "iter": 350, "lr": 0.02, "time": 0.5269, "data_time": 0.01887, "memory": 5696, "loss_rpn_cls": 0.01354, "loss_rpn_bbox": 0.00632, "loss_cls": 0.13872, "acc": 95.76758, "loss_bbox": 0.0718, "loss": 0.23038} 42 | {"mode": "train", "epoch": 4, "iter": 400, "lr": 0.02, "time": 0.51912, "data_time": 0.01523, "memory": 5696, "loss_rpn_cls": 0.0145, "loss_rpn_bbox": 0.00734, "loss_cls": 0.14204, "acc": 95.16992, "loss_bbox": 0.07278, "loss": 0.23666} 43 | {"mode": "train", "epoch": 4, "iter": 450, "lr": 0.02, "time": 1.12456, "data_time": 0.6056, "memory": 5696, "loss_rpn_cls": 0.01776, "loss_rpn_bbox": 0.00924, "loss_cls": 0.13795, "acc": 95.61914, "loss_bbox": 0.07037, "loss": 0.23531} 44 | {"mode": "train", "epoch": 4, "iter": 500, "lr": 0.02, "time": 0.51186, "data_time": 0.01212, "memory": 5696, "loss_rpn_cls": 0.01487, "loss_rpn_bbox": 0.00719, "loss_cls": 0.13872, "acc": 95.75977, "loss_bbox": 0.07414, "loss": 0.23492} 45 | {"mode": "train", "epoch": 4, "iter": 550, "lr": 0.02, "time": 0.50763, "data_time": 0.00727, "memory": 5696, "loss_rpn_cls": 0.011, "loss_rpn_bbox": 0.00708, "loss_cls": 0.12878, "acc": 95.68555, "loss_bbox": 0.06927, "loss": 0.21613} 46 | {"mode": "train", "epoch": 5, "iter": 115, "lr": 0.02, "time": 0.43991, "data_time": 0.21911, "memory": 5696, "loss_rpn_cls": 0.01058, "loss_rpn_bbox": 0.00791, "loss_cls": 0.15151, "acc": 94.80554, "loss_bbox": 0.07569, "loss": 0.24569} 47 | {"mode": "train", "epoch": 5, "iter": 50, "lr": 0.02, "time": 0.52744, "data_time": 0.02597, "memory": 5696, "loss_rpn_cls": 0.01129, "loss_rpn_bbox": 0.00606, "loss_cls": 0.13429, "acc": 95.52344, "loss_bbox": 0.065, "loss": 0.21663} 48 | {"mode": "train", "epoch": 5, "iter": 100, "lr": 0.02, "time": 0.50221, "data_time": 0.00637, "memory": 5696, "loss_rpn_cls": 0.01236, "loss_rpn_bbox": 0.0061, "loss_cls": 0.13439, "acc": 95.18945, "loss_bbox": 0.06321, "loss": 0.21608} 49 | {"mode": "train", "epoch": 5, "iter": 150, "lr": 0.02, "time": 0.50715, "data_time": 0.00642, "memory": 5696, "loss_rpn_cls": 0.01211, "loss_rpn_bbox": 0.00776, "loss_cls": 0.14989, "acc": 94.78711, "loss_bbox": 0.07005, "loss": 0.2398} 50 | {"mode": "train", "epoch": 5, "iter": 200, "lr": 0.02, "time": 0.50427, "data_time": 0.00638, "memory": 5696, "loss_rpn_cls": 0.00914, "loss_rpn_bbox": 0.00716, "loss_cls": 0.13213, "acc": 95.38672, "loss_bbox": 0.06545, "loss": 0.21388} 51 | {"mode": "train", "epoch": 5, "iter": 250, "lr": 0.02, "time": 0.53265, "data_time": 0.01709, "memory": 5696, "loss_rpn_cls": 0.01041, "loss_rpn_bbox": 0.00724, "loss_cls": 0.13253, "acc": 95.34375, "loss_bbox": 0.06543, "loss": 0.21561} 52 | {"mode": "train", "epoch": 5, "iter": 300, "lr": 0.02, "time": 0.50917, "data_time": 0.00716, "memory": 5696, "loss_rpn_cls": 0.01345, "loss_rpn_bbox": 0.00737, "loss_cls": 0.13211, "acc": 95.46484, "loss_bbox": 0.06506, "loss": 0.21799} 53 | {"mode": "train", "epoch": 5, "iter": 350, "lr": 0.02, "time": 0.52442, "data_time": 0.01546, "memory": 5696, "loss_rpn_cls": 0.01018, "loss_rpn_bbox": 0.00816, "loss_cls": 0.12542, "acc": 95.55273, "loss_bbox": 0.05815, "loss": 0.20191} 54 | {"mode": "train", "epoch": 5, "iter": 400, "lr": 0.02, "time": 0.54248, "data_time": 0.04472, "memory": 5696, "loss_rpn_cls": 0.01182, "loss_rpn_bbox": 0.00879, "loss_cls": 0.13112, "acc": 95.25977, "loss_bbox": 0.05756, "loss": 0.20928} 55 | {"mode": "train", "epoch": 5, "iter": 450, "lr": 0.02, "time": 0.54863, "data_time": 0.03904, "memory": 5696, "loss_rpn_cls": 0.01463, "loss_rpn_bbox": 0.00769, "loss_cls": 0.15927, "acc": 94.65625, "loss_bbox": 0.07346, "loss": 0.25505} 56 | {"mode": "train", "epoch": 5, "iter": 500, "lr": 0.02, "time": 0.61727, "data_time": 0.03714, "memory": 5696, "loss_rpn_cls": 0.01238, "loss_rpn_bbox": 0.00644, "loss_cls": 0.14063, "acc": 95.21875, "loss_bbox": 0.06167, "loss": 0.22113} 57 | {"mode": "train", "epoch": 5, "iter": 550, "lr": 0.02, "time": 0.55806, "data_time": 0.00825, "memory": 5696, "loss_rpn_cls": 0.01484, "loss_rpn_bbox": 0.00729, "loss_cls": 0.13206, "acc": 95.5957, "loss_bbox": 0.06198, "loss": 0.21617} 58 | {"mode": "train", "epoch": 6, "iter": 50, "lr": 0.02, "time": 0.54323, "data_time": 0.04421, "memory": 5696, "loss_rpn_cls": 0.00791, "loss_rpn_bbox": 0.00557, "loss_cls": 0.12604, "acc": 95.40039, "loss_bbox": 0.06172, "loss": 0.20124} 59 | {"mode": "train", "epoch": 6, "iter": 100, "lr": 0.02, "time": 0.69622, "data_time": 0.19796, "memory": 5696, "loss_rpn_cls": 0.00921, "loss_rpn_bbox": 0.00666, "loss_cls": 0.13228, "acc": 95.25, "loss_bbox": 0.0594, "loss": 0.20755} 60 | {"mode": "train", "epoch": 6, "iter": 150, "lr": 0.02, "time": 0.60573, "data_time": 0.04169, "memory": 5696, "loss_rpn_cls": 0.01445, "loss_rpn_bbox": 0.0085, "loss_cls": 0.15071, "acc": 94.96484, "loss_bbox": 0.06803, "loss": 0.2417} 61 | {"mode": "train", "epoch": 6, "iter": 200, "lr": 0.02, "time": 0.52496, "data_time": 0.0168, "memory": 5696, "loss_rpn_cls": 0.00737, "loss_rpn_bbox": 0.00577, "loss_cls": 0.12484, "acc": 95.45898, "loss_bbox": 0.06018, "loss": 0.19816} 62 | {"mode": "train", "epoch": 6, "iter": 250, "lr": 0.02, "time": 0.55447, "data_time": 0.04938, "memory": 5696, "loss_rpn_cls": 0.01032, "loss_rpn_bbox": 0.00713, "loss_cls": 0.12543, "acc": 95.29492, "loss_bbox": 0.0574, "loss": 0.20028} 63 | {"mode": "train", "epoch": 6, "iter": 300, "lr": 0.02, "time": 0.52515, "data_time": 0.02216, "memory": 5696, "loss_rpn_cls": 0.01372, "loss_rpn_bbox": 0.00662, "loss_cls": 0.12372, "acc": 95.66602, "loss_bbox": 0.06285, "loss": 0.20691} 64 | {"mode": "train", "epoch": 6, "iter": 350, "lr": 0.02, "time": 0.52318, "data_time": 0.02102, "memory": 5696, "loss_rpn_cls": 0.00815, "loss_rpn_bbox": 0.00604, "loss_cls": 0.14505, "acc": 94.71289, "loss_bbox": 0.06095, "loss": 0.22018} 65 | {"mode": "train", "epoch": 6, "iter": 400, "lr": 0.02, "time": 0.51061, "data_time": 0.00707, "memory": 5696, "loss_rpn_cls": 0.00853, "loss_rpn_bbox": 0.00593, "loss_cls": 0.11872, "acc": 95.63086, "loss_bbox": 0.05667, "loss": 0.18986} 66 | {"mode": "train", "epoch": 6, "iter": 450, "lr": 0.02, "time": 0.50287, "data_time": 0.00686, "memory": 5696, "loss_rpn_cls": 0.00889, "loss_rpn_bbox": 0.00609, "loss_cls": 0.13925, "acc": 95.15234, "loss_bbox": 0.06055, "loss": 0.21478} 67 | {"mode": "train", "epoch": 6, "iter": 500, "lr": 0.02, "time": 0.50468, "data_time": 0.00674, "memory": 5696, "loss_rpn_cls": 0.00943, "loss_rpn_bbox": 0.00637, "loss_cls": 0.133, "acc": 95.20898, "loss_bbox": 0.05548, "loss": 0.20428} 68 | {"mode": "train", "epoch": 6, "iter": 550, "lr": 0.02, "time": 0.5879, "data_time": 0.08472, "memory": 5696, "loss_rpn_cls": 0.01385, "loss_rpn_bbox": 0.00727, "loss_cls": 0.11971, "acc": 95.67578, "loss_bbox": 0.05216, "loss": 0.19299} 69 | {"mode": "train", "epoch": 7, "iter": 115, "lr": 0.02, "time": 0.51148, "data_time": 0.27775, "memory": 5696, "loss_rpn_cls": 0.00773, "loss_rpn_bbox": 0.00594, "loss_cls": 0.12132, "acc": 95.58169, "loss_bbox": 0.05579, "loss": 0.19078} 70 | {"mode": "train", "epoch": 7, "iter": 50, "lr": 0.02, "time": 0.60383, "data_time": 0.03955, "memory": 5696, "loss_rpn_cls": 0.0114, "loss_rpn_bbox": 0.00631, "loss_cls": 0.1394, "acc": 95.04297, "loss_bbox": 0.05741, "loss": 0.21452} 71 | {"mode": "train", "epoch": 7, "iter": 100, "lr": 0.02, "time": 0.50432, "data_time": 0.00607, "memory": 5696, "loss_rpn_cls": 0.01491, "loss_rpn_bbox": 0.00806, "loss_cls": 0.15647, "acc": 94.87305, "loss_bbox": 0.06381, "loss": 0.24324} 72 | {"mode": "train", "epoch": 7, "iter": 150, "lr": 0.02, "time": 0.71423, "data_time": 0.14432, "memory": 5696, "loss_rpn_cls": 0.03039, "loss_rpn_bbox": 0.01024, "loss_cls": 0.15473, "acc": 94.99414, "loss_bbox": 0.07707, "loss": 0.27241} 73 | {"mode": "train", "epoch": 7, "iter": 200, "lr": 0.02, "time": 0.72053, "data_time": 0.10828, "memory": 5696, "loss_rpn_cls": 0.01012, "loss_rpn_bbox": 0.00624, "loss_cls": 0.1124, "acc": 96.11914, "loss_bbox": 0.05746, "loss": 0.18621} 74 | {"mode": "train", "epoch": 7, "iter": 250, "lr": 0.02, "time": 0.6184, "data_time": 0.00939, "memory": 5696, "loss_rpn_cls": 0.00973, "loss_rpn_bbox": 0.00509, "loss_cls": 0.11366, "acc": 96.02734, "loss_bbox": 0.04682, "loss": 0.17529} 75 | {"mode": "train", "epoch": 7, "iter": 300, "lr": 0.02, "time": 0.61433, "data_time": 0.01004, "memory": 5696, "loss_rpn_cls": 0.01186, "loss_rpn_bbox": 0.00697, "loss_cls": 0.1158, "acc": 95.9668, "loss_bbox": 0.05012, "loss": 0.18476} 76 | {"mode": "train", "epoch": 7, "iter": 350, "lr": 0.02, "time": 0.62333, "data_time": 0.00981, "memory": 5696, "loss_rpn_cls": 0.00889, "loss_rpn_bbox": 0.00581, "loss_cls": 0.11227, "acc": 95.95898, "loss_bbox": 0.04713, "loss": 0.1741} 77 | {"mode": "train", "epoch": 7, "iter": 400, "lr": 0.02, "time": 0.63422, "data_time": 0.02249, "memory": 5696, "loss_rpn_cls": 0.00838, "loss_rpn_bbox": 0.0056, "loss_cls": 0.10986, "acc": 95.9375, "loss_bbox": 0.04764, "loss": 0.17148} 78 | {"mode": "train", "epoch": 7, "iter": 450, "lr": 0.02, "time": 0.62604, "data_time": 0.00966, "memory": 5696, "loss_rpn_cls": 0.00623, "loss_rpn_bbox": 0.0058, "loss_cls": 0.1114, "acc": 95.74609, "loss_bbox": 0.05365, "loss": 0.17709} 79 | {"mode": "train", "epoch": 7, "iter": 500, "lr": 0.02, "time": 0.66864, "data_time": 0.05844, "memory": 5696, "loss_rpn_cls": 0.00848, "loss_rpn_bbox": 0.00699, "loss_cls": 0.13194, "acc": 95.1543, "loss_bbox": 0.05894, "loss": 0.20636} 80 | {"mode": "train", "epoch": 7, "iter": 550, "lr": 0.02, "time": 0.61844, "data_time": 0.00958, "memory": 5696, "loss_rpn_cls": 0.00906, "loss_rpn_bbox": 0.00752, "loss_cls": 0.14079, "acc": 95.05859, "loss_bbox": 0.06675, "loss": 0.22413} 81 | {"mode": "train", "epoch": 8, "iter": 50, "lr": 0.02, "time": 0.53554, "data_time": 0.02993, "memory": 5696, "loss_rpn_cls": 0.01006, "loss_rpn_bbox": 0.00743, "loss_cls": 0.12461, "acc": 95.3457, "loss_bbox": 0.05612, "loss": 0.19822} 82 | {"mode": "train", "epoch": 8, "iter": 100, "lr": 0.02, "time": 0.50579, "data_time": 0.00687, "memory": 5696, "loss_rpn_cls": 0.00728, "loss_rpn_bbox": 0.00698, "loss_cls": 0.11898, "acc": 95.42578, "loss_bbox": 0.05315, "loss": 0.18638} 83 | {"mode": "train", "epoch": 8, "iter": 150, "lr": 0.02, "time": 0.50781, "data_time": 0.00655, "memory": 5696, "loss_rpn_cls": 0.00811, "loss_rpn_bbox": 0.0063, "loss_cls": 0.11761, "acc": 95.57422, "loss_bbox": 0.05272, "loss": 0.18474} 84 | {"mode": "train", "epoch": 8, "iter": 200, "lr": 0.02, "time": 0.50765, "data_time": 0.00667, "memory": 5696, "loss_rpn_cls": 0.00802, "loss_rpn_bbox": 0.00556, "loss_cls": 0.12714, "acc": 95.42773, "loss_bbox": 0.05261, "loss": 0.19334} 85 | {"mode": "train", "epoch": 8, "iter": 250, "lr": 0.02, "time": 0.52218, "data_time": 0.01573, "memory": 5696, "loss_rpn_cls": 0.00745, "loss_rpn_bbox": 0.00507, "loss_cls": 0.10948, "acc": 96.04492, "loss_bbox": 0.04383, "loss": 0.16583} 86 | {"mode": "train", "epoch": 8, "iter": 300, "lr": 0.02, "time": 0.544, "data_time": 0.03903, "memory": 5696, "loss_rpn_cls": 0.0073, "loss_rpn_bbox": 0.00615, "loss_cls": 0.11153, "acc": 95.77148, "loss_bbox": 0.04881, "loss": 0.17379} 87 | {"mode": "train", "epoch": 8, "iter": 350, "lr": 0.02, "time": 1.09267, "data_time": 0.59414, "memory": 5696, "loss_rpn_cls": 0.00766, "loss_rpn_bbox": 0.00582, "loss_cls": 0.11566, "acc": 95.80273, "loss_bbox": 0.05777, "loss": 0.18691} 88 | {"mode": "train", "epoch": 8, "iter": 400, "lr": 0.02, "time": 0.50323, "data_time": 0.00671, "memory": 5696, "loss_rpn_cls": 0.01412, "loss_rpn_bbox": 0.00728, "loss_cls": 0.11361, "acc": 96.06641, "loss_bbox": 0.05286, "loss": 0.18787} 89 | {"mode": "train", "epoch": 8, "iter": 450, "lr": 0.02, "time": 0.51021, "data_time": 0.00673, "memory": 5696, "loss_rpn_cls": 0.01137, "loss_rpn_bbox": 0.00646, "loss_cls": 0.11739, "acc": 95.77734, "loss_bbox": 0.05152, "loss": 0.18675} 90 | {"mode": "train", "epoch": 8, "iter": 500, "lr": 0.02, "time": 0.51532, "data_time": 0.01624, "memory": 5696, "loss_rpn_cls": 0.00803, "loss_rpn_bbox": 0.00589, "loss_cls": 0.09256, "acc": 96.40625, "loss_bbox": 0.04921, "loss": 0.15568} 91 | {"mode": "train", "epoch": 8, "iter": 550, "lr": 0.02, "time": 0.50599, "data_time": 0.00654, "memory": 5696, "loss_rpn_cls": 0.00736, "loss_rpn_bbox": 0.00455, "loss_cls": 0.10456, "acc": 95.81836, "loss_bbox": 0.04204, "loss": 0.15851} 92 | {"mode": "train", "epoch": 9, "iter": 115, "lr": 0.02, "time": 0.47793, "data_time": 0.21516, "memory": 5696, "loss_rpn_cls": 0.00941, "loss_rpn_bbox": 0.00556, "loss_cls": 0.10351, "acc": 96.1464, "loss_bbox": 0.05045, "loss": 0.16893} 93 | {"mode": "train", "epoch": 9, "iter": 50, "lr": 0.002, "time": 0.66657, "data_time": 0.0535, "memory": 5696, "loss_rpn_cls": 0.00783, "loss_rpn_bbox": 0.00554, "loss_cls": 0.09764, "acc": 96.47461, "loss_bbox": 0.04826, "loss": 0.15928} 94 | {"mode": "train", "epoch": 9, "iter": 100, "lr": 0.002, "time": 0.62643, "data_time": 0.01115, "memory": 5696, "loss_rpn_cls": 0.0112, "loss_rpn_bbox": 0.00533, "loss_cls": 0.10305, "acc": 96.04688, "loss_bbox": 0.04397, "loss": 0.16356} 95 | {"mode": "train", "epoch": 9, "iter": 150, "lr": 0.002, "time": 0.62425, "data_time": 0.00944, "memory": 5696, "loss_rpn_cls": 0.00628, "loss_rpn_bbox": 0.00489, "loss_cls": 0.08919, "acc": 96.45508, "loss_bbox": 0.03748, "loss": 0.13783} 96 | {"mode": "train", "epoch": 9, "iter": 200, "lr": 0.002, "time": 0.62269, "data_time": 0.00947, "memory": 5696, "loss_rpn_cls": 0.00727, "loss_rpn_bbox": 0.00399, "loss_cls": 0.09384, "acc": 96.51758, "loss_bbox": 0.03587, "loss": 0.14097} 97 | {"mode": "train", "epoch": 9, "iter": 250, "lr": 0.002, "time": 0.62472, "data_time": 0.0096, "memory": 5696, "loss_rpn_cls": 0.00641, "loss_rpn_bbox": 0.00699, "loss_cls": 0.10836, "acc": 95.74609, "loss_bbox": 0.05005, "loss": 0.1718} 98 | {"mode": "train", "epoch": 9, "iter": 300, "lr": 0.002, "time": 0.61771, "data_time": 0.00952, "memory": 5696, "loss_rpn_cls": 0.00587, "loss_rpn_bbox": 0.00415, "loss_cls": 0.08779, "acc": 96.65039, "loss_bbox": 0.03464, "loss": 0.13246} 99 | {"mode": "train", "epoch": 9, "iter": 350, "lr": 0.002, "time": 0.61732, "data_time": 0.00954, "memory": 5696, "loss_rpn_cls": 0.00727, "loss_rpn_bbox": 0.00493, "loss_cls": 0.09233, "acc": 96.36133, "loss_bbox": 0.03873, "loss": 0.14327} 100 | {"mode": "train", "epoch": 9, "iter": 400, "lr": 0.002, "time": 0.74154, "data_time": 0.13262, "memory": 5696, "loss_rpn_cls": 0.00466, "loss_rpn_bbox": 0.00375, "loss_cls": 0.09513, "acc": 96.31641, "loss_bbox": 0.0429, "loss": 0.14644} 101 | {"mode": "train", "epoch": 9, "iter": 450, "lr": 0.002, "time": 0.62911, "data_time": 0.01007, "memory": 5696, "loss_rpn_cls": 0.00534, "loss_rpn_bbox": 0.00411, "loss_cls": 0.08636, "acc": 96.52539, "loss_bbox": 0.03979, "loss": 0.13559} 102 | {"mode": "train", "epoch": 9, "iter": 500, "lr": 0.002, "time": 0.57786, "data_time": 0.00847, "memory": 5696, "loss_rpn_cls": 0.00463, "loss_rpn_bbox": 0.00363, "loss_cls": 0.08264, "acc": 96.76953, "loss_bbox": 0.03445, "loss": 0.12536} 103 | {"mode": "train", "epoch": 9, "iter": 550, "lr": 0.002, "time": 0.50238, "data_time": 0.00635, "memory": 5696, "loss_rpn_cls": 0.00399, "loss_rpn_bbox": 0.00363, "loss_cls": 0.07193, "acc": 97.25781, "loss_bbox": 0.02702, "loss": 0.10657} 104 | {"mode": "train", "epoch": 10, "iter": 50, "lr": 0.002, "time": 0.65076, "data_time": 0.03772, "memory": 5696, "loss_rpn_cls": 0.00544, "loss_rpn_bbox": 0.0048, "loss_cls": 0.088, "acc": 96.51953, "loss_bbox": 0.03588, "loss": 0.13412} 105 | {"mode": "train", "epoch": 10, "iter": 100, "lr": 0.002, "time": 0.62092, "data_time": 0.00971, "memory": 5696, "loss_rpn_cls": 0.00474, "loss_rpn_bbox": 0.00415, "loss_cls": 0.07196, "acc": 97.14062, "loss_bbox": 0.02935, "loss": 0.1102} 106 | {"mode": "train", "epoch": 10, "iter": 150, "lr": 0.002, "time": 0.61863, "data_time": 0.00968, "memory": 5696, "loss_rpn_cls": 0.00706, "loss_rpn_bbox": 0.0052, "loss_cls": 0.10249, "acc": 96.06055, "loss_bbox": 0.03915, "loss": 0.1539} 107 | {"mode": "train", "epoch": 10, "iter": 200, "lr": 0.002, "time": 0.61971, "data_time": 0.00964, "memory": 5696, "loss_rpn_cls": 0.00707, "loss_rpn_bbox": 0.00481, "loss_cls": 0.08109, "acc": 97.04102, "loss_bbox": 0.03655, "loss": 0.12951} 108 | {"mode": "train", "epoch": 10, "iter": 250, "lr": 0.002, "time": 0.65088, "data_time": 0.04016, "memory": 5696, "loss_rpn_cls": 0.00403, "loss_rpn_bbox": 0.00388, "loss_cls": 0.06913, "acc": 97.375, "loss_bbox": 0.03145, "loss": 0.10849} 109 | {"mode": "train", "epoch": 10, "iter": 300, "lr": 0.002, "time": 0.62276, "data_time": 0.00966, "memory": 5696, "loss_rpn_cls": 0.00469, "loss_rpn_bbox": 0.00388, "loss_cls": 0.08334, "acc": 96.84961, "loss_bbox": 0.03638, "loss": 0.12829} 110 | {"mode": "train", "epoch": 10, "iter": 350, "lr": 0.002, "time": 0.62188, "data_time": 0.01001, "memory": 5696, "loss_rpn_cls": 0.00551, "loss_rpn_bbox": 0.00414, "loss_cls": 0.07683, "acc": 96.92969, "loss_bbox": 0.0334, "loss": 0.11988} 111 | {"mode": "train", "epoch": 10, "iter": 400, "lr": 0.002, "time": 0.62571, "data_time": 0.00958, "memory": 5696, "loss_rpn_cls": 0.00509, "loss_rpn_bbox": 0.00395, "loss_cls": 0.08428, "acc": 96.5293, "loss_bbox": 0.0345, "loss": 0.12781} 112 | {"mode": "train", "epoch": 10, "iter": 450, "lr": 0.002, "time": 0.62705, "data_time": 0.00964, "memory": 5696, "loss_rpn_cls": 0.00391, "loss_rpn_bbox": 0.00393, "loss_cls": 0.06881, "acc": 97.16602, "loss_bbox": 0.03043, "loss": 0.10708} 113 | {"mode": "train", "epoch": 10, "iter": 500, "lr": 0.002, "time": 0.62267, "data_time": 0.00957, "memory": 5696, "loss_rpn_cls": 0.0062, "loss_rpn_bbox": 0.00457, "loss_cls": 0.09365, "acc": 96.18359, "loss_bbox": 0.03698, "loss": 0.14141} 114 | {"mode": "train", "epoch": 10, "iter": 550, "lr": 0.002, "time": 0.62301, "data_time": 0.00966, "memory": 5696, "loss_rpn_cls": 0.00408, "loss_rpn_bbox": 0.00397, "loss_cls": 0.08208, "acc": 96.70312, "loss_bbox": 0.03618, "loss": 0.12632} 115 | {"mode": "train", "epoch": 11, "iter": 115, "lr": 0.002, "time": 0.46758, "data_time": 0.18141, "memory": 5696, "loss_rpn_cls": 0.00526, "loss_rpn_bbox": 0.00422, "loss_cls": 0.07642, "acc": 97.09834, "loss_bbox": 0.03318, "loss": 0.11907} 116 | {"mode": "train", "epoch": 11, "iter": 50, "lr": 0.002, "time": 0.63706, "data_time": 0.0285, "memory": 5696, "loss_rpn_cls": 0.00585, "loss_rpn_bbox": 0.00429, "loss_cls": 0.07606, "acc": 96.89453, "loss_bbox": 0.03286, "loss": 0.11906} 117 | {"mode": "train", "epoch": 11, "iter": 100, "lr": 0.002, "time": 0.54277, "data_time": 0.00764, "memory": 5696, "loss_rpn_cls": 0.0049, "loss_rpn_bbox": 0.00448, "loss_cls": 0.08597, "acc": 96.70117, "loss_bbox": 0.03729, "loss": 0.13264} 118 | {"mode": "train", "epoch": 11, "iter": 150, "lr": 0.002, "time": 0.51208, "data_time": 0.00926, "memory": 5696, "loss_rpn_cls": 0.00422, "loss_rpn_bbox": 0.00381, "loss_cls": 0.07464, "acc": 97.12109, "loss_bbox": 0.03192, "loss": 0.11459} 119 | {"mode": "train", "epoch": 11, "iter": 200, "lr": 0.002, "time": 0.50722, "data_time": 0.00935, "memory": 5696, "loss_rpn_cls": 0.00586, "loss_rpn_bbox": 0.00424, "loss_cls": 0.07833, "acc": 97.02344, "loss_bbox": 0.03441, "loss": 0.12285} 120 | {"mode": "train", "epoch": 11, "iter": 250, "lr": 0.002, "time": 0.56258, "data_time": 0.06363, "memory": 5696, "loss_rpn_cls": 0.00562, "loss_rpn_bbox": 0.00477, "loss_cls": 0.08335, "acc": 96.95898, "loss_bbox": 0.03518, "loss": 0.12891} 121 | {"mode": "train", "epoch": 11, "iter": 300, "lr": 0.002, "time": 0.51535, "data_time": 0.01336, "memory": 5696, "loss_rpn_cls": 0.0047, "loss_rpn_bbox": 0.00403, "loss_cls": 0.07576, "acc": 97.0918, "loss_bbox": 0.0347, "loss": 0.1192} 122 | {"mode": "train", "epoch": 11, "iter": 350, "lr": 0.002, "time": 0.70907, "data_time": 0.1187, "memory": 5696, "loss_rpn_cls": 0.00535, "loss_rpn_bbox": 0.00338, "loss_cls": 0.0737, "acc": 97.31836, "loss_bbox": 0.03076, "loss": 0.11319} 123 | {"mode": "train", "epoch": 11, "iter": 400, "lr": 0.002, "time": 1.27861, "data_time": 0.67146, "memory": 5696, "loss_rpn_cls": 0.0037, "loss_rpn_bbox": 0.00346, "loss_cls": 0.07627, "acc": 96.89453, "loss_bbox": 0.02839, "loss": 0.11182} 124 | {"mode": "train", "epoch": 11, "iter": 450, "lr": 0.002, "time": 0.5671, "data_time": 0.00835, "memory": 5696, "loss_rpn_cls": 0.00387, "loss_rpn_bbox": 0.004, "loss_cls": 0.06613, "acc": 97.42969, "loss_bbox": 0.03056, "loss": 0.10455} 125 | {"mode": "train", "epoch": 11, "iter": 500, "lr": 0.002, "time": 0.60135, "data_time": 0.09074, "memory": 5696, "loss_rpn_cls": 0.00402, "loss_rpn_bbox": 0.0038, "loss_cls": 0.07879, "acc": 96.77734, "loss_bbox": 0.032, "loss": 0.11861} 126 | {"mode": "train", "epoch": 11, "iter": 550, "lr": 0.002, "time": 0.71062, "data_time": 0.16395, "memory": 5696, "loss_rpn_cls": 0.00457, "loss_rpn_bbox": 0.00421, "loss_cls": 0.07552, "acc": 97.07227, "loss_bbox": 0.03425, "loss": 0.11855} 127 | {"mode": "train", "epoch": 12, "iter": 50, "lr": 0.0002, "time": 0.68538, "data_time": 0.07949, "memory": 5696, "loss_rpn_cls": 0.00303, "loss_rpn_bbox": 0.00349, "loss_cls": 0.06376, "acc": 97.4668, "loss_bbox": 0.02717, "loss": 0.09745} 128 | {"mode": "train", "epoch": 12, "iter": 100, "lr": 0.0002, "time": 0.52017, "data_time": 0.0141, "memory": 5696, "loss_rpn_cls": 0.00689, "loss_rpn_bbox": 0.0041, "loss_cls": 0.06425, "acc": 97.46484, "loss_bbox": 0.026, "loss": 0.10124} 129 | {"mode": "train", "epoch": 12, "iter": 150, "lr": 0.0002, "time": 0.52325, "data_time": 0.02226, "memory": 5696, "loss_rpn_cls": 0.00297, "loss_rpn_bbox": 0.00397, "loss_cls": 0.07261, "acc": 97.19531, "loss_bbox": 0.0298, "loss": 0.10935} 130 | {"mode": "train", "epoch": 12, "iter": 200, "lr": 0.0002, "time": 0.51168, "data_time": 0.00675, "memory": 5696, "loss_rpn_cls": 0.00506, "loss_rpn_bbox": 0.00398, "loss_cls": 0.07436, "acc": 97.12109, "loss_bbox": 0.03109, "loss": 0.11448} 131 | {"mode": "train", "epoch": 12, "iter": 250, "lr": 0.0002, "time": 0.56226, "data_time": 0.05819, "memory": 5696, "loss_rpn_cls": 0.00368, "loss_rpn_bbox": 0.00398, "loss_cls": 0.07391, "acc": 97.19336, "loss_bbox": 0.03348, "loss": 0.11504} 132 | {"mode": "train", "epoch": 12, "iter": 300, "lr": 0.0002, "time": 0.50746, "data_time": 0.00649, "memory": 5696, "loss_rpn_cls": 0.00489, "loss_rpn_bbox": 0.00389, "loss_cls": 0.07403, "acc": 97.19141, "loss_bbox": 0.03107, "loss": 0.11388} 133 | {"mode": "train", "epoch": 12, "iter": 350, "lr": 0.0002, "time": 0.68346, "data_time": 0.17951, "memory": 5696, "loss_rpn_cls": 0.00464, "loss_rpn_bbox": 0.00429, "loss_cls": 0.07448, "acc": 97.33398, "loss_bbox": 0.03567, "loss": 0.11909} 134 | {"mode": "train", "epoch": 12, "iter": 400, "lr": 0.0002, "time": 0.57773, "data_time": 0.04374, "memory": 5696, "loss_rpn_cls": 0.00353, "loss_rpn_bbox": 0.00412, "loss_cls": 0.07164, "acc": 97.17383, "loss_bbox": 0.03015, "loss": 0.10944} 135 | {"mode": "train", "epoch": 12, "iter": 450, "lr": 0.0002, "time": 0.52545, "data_time": 0.0215, "memory": 5696, "loss_rpn_cls": 0.00539, "loss_rpn_bbox": 0.00419, "loss_cls": 0.07324, "acc": 97.15234, "loss_bbox": 0.03333, "loss": 0.11615} 136 | {"mode": "train", "epoch": 12, "iter": 500, "lr": 0.0002, "time": 0.53737, "data_time": 0.03108, "memory": 5696, "loss_rpn_cls": 0.0038, "loss_rpn_bbox": 0.00354, "loss_cls": 0.06582, "acc": 97.54492, "loss_bbox": 0.03, "loss": 0.10316} 137 | {"mode": "train", "epoch": 12, "iter": 550, "lr": 0.0002, "time": 0.51086, "data_time": 0.01335, "memory": 5696, "loss_rpn_cls": 0.0042, "loss_rpn_bbox": 0.00334, "loss_cls": 0.06598, "acc": 97.62109, "loss_bbox": 0.02652, "loss": 0.10004} 138 | {"mode": "train", "epoch": 13, "iter": 115, "lr": 0.0002, "time": 0.48316, "data_time": 0.26549, "memory": 5696, "loss_rpn_cls": 0.00433, "loss_rpn_bbox": 0.00392, "loss_cls": 0.0703, "acc": 97.23336, "loss_bbox": 0.03131, "loss": 0.10985} --------------------------------------------------------------------------------