# 3D-ResNets-PaddlePaddle **Repository Path**: KevinPang180_admin/PaddlePaddle_3D_ResNets ## Basic Information - **Project Name**: 3D-ResNets-PaddlePaddle - **Description**: 论文复现 - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2020-09-06 - **Last Updated**: 2020-12-19 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # 3D-ResNets-PaddlePaddle 通过使用PaddlePaddle复现3D ResNets论文 # 1 前期准备工作 在使用PaddlePaddle复现论文代码前，我们需要先理解3D_ResNets论文并用Pytorch跑通一遍代码，理解其传入参数、网络结构、传出参数等。这些我就统一归为前期准备工作。 以下是本人先前阅读论文《Learning Spatio-Temporal Features with 3D Residual Networks for Action Recognition》时做的一些笔记： https://blog.csdn.net/qq_41138197/article/details/107832207 以下是论文《Would Mega-scale Datasets Further Enhance Spatiotemporal 3D CNNs》的下载地址： https://arxiv.org/abs/2004.04968 ## 1.1 本地环境配置 - 首先，需要一台N卡的笔记本，然后配置好CUDA和cudnn。Linux环境下Arch系配置方式可参考https://www.bilibili.com/read/cv6438916?spm_id_from=333.788.b_636f6d6d656e74.7 - 然后，安装paddlepaddle-gpu和pytorch。（PaddlePaddle安装可参考https://www.paddlepaddle.org.cn/install/quick；Pytorch建议直接用pip安装pip install torch torchvision） - 运行过程中如出现"No moudle..."，一般情况下表明你需要使用pip安装对应的包，或者需要调整文件目录结构以保证能够正确调用本地的py文件。 ### 本地CUDA安装截图 配置方式可参考：https://www.bilibili.com/read/cv6438916?spm_id_from=333.788.b_636f6d6d656e74.7 ![image](pic/cuda环境.png) ## 1.2 目录结构设置 **目录结构**：可理解为文件层级目录，无论是本地还是远程服务器都需要按论文中的要求配置。 ``` ~/ data/ kinetics_videos/ jpg/ .../ (directories of class names) .../ (directories of video names) ... (jpg files) results/ save_100.pth kinetics.json ``` ## 1.3 理解main.py参数 在执行python main.py时，后可跟一系列的参数，它们是什么意思？有什么作用？这些我们需要在训练前就先尽可能有一个相对清晰的概念。 进入GitHub源码解压后的文件目录，输入 `python main.py -h` 即可查看，若前面配置有小伙伴无法满足（比如N卡），可参考我在本地查到的help： ![imgame](pic/main.py参数设置.png) ```bash [kevinpang@MateBookProX ~]$ cd /home/kevinpang/Documents/CodeLibrary/3D-ResNet/Learning-Spatio-Temporal-Features-with-3D-ResNets-PyTorch [kevinpang@MateBookProX Learning-Spatio-Temporal-Features-with-3D-ResNets-PyTorch]$ python main.py -h usage: main.py [-h] [--root_path ROOT_PATH] [--video_path VIDEO_PATH] [--annotation_path ANNOTATION_PATH] [--result_path RESULT_PATH] [--dataset DATASET] [--n_classes N_CLASSES] [--n_pretrain_classes N_PRETRAIN_CLASSES] [--pretrain_path PRETRAIN_PATH] [--ft_begin_module FT_BEGIN_MODULE] [--sample_size SAMPLE_SIZE] [--sample_duration SAMPLE_DURATION] [--sample_t_stride SAMPLE_T_STRIDE] [--train_crop TRAIN_CROP] [--train_crop_min_scale TRAIN_CROP_MIN_SCALE] [--train_crop_min_ratio TRAIN_CROP_MIN_RATIO] [--no_hflip] [--colorjitter] [--train_t_crop TRAIN_T_CROP] [--learning_rate LEARNING_RATE] [--momentum MOMENTUM] [--dampening DAMPENING] [--weight_decay WEIGHT_DECAY] [--mean_dataset MEAN_DATASET] [--no_mean_norm] [--no_std_norm] [--value_scale VALUE_SCALE] [--nesterov] [--optimizer OPTIMIZER] [--lr_scheduler LR_SCHEDULER] [--multistep_milestones MULTISTEP_MILESTONES [MULTISTEP_MILESTONES ...]] [--overwrite_milestones] [--plateau_patience PLATEAU_PATIENCE] [--batch_size BATCH_SIZE] [--inference_batch_size INFERENCE_BATCH_SIZE] [--batchnorm_sync] [--n_epochs N_EPOCHS] [--n_val_samples N_VAL_SAMPLES] [--resume_path RESUME_PATH] [--no_train] [--no_val] [--inference] [--inference_subset INFERENCE_SUBSET] [--inference_stride INFERENCE_STRIDE] [--inference_crop INFERENCE_CROP] [--inference_no_average] [--no_cuda] [--n_threads N_THREADS] [--checkpoint CHECKPOINT] [--model MODEL] [--model_depth MODEL_DEPTH] [--conv1_t_size CONV1_T_SIZE] [--conv1_t_stride CONV1_T_STRIDE] [--no_max_pool] [--resnet_shortcut RESNET_SHORTCUT] [--resnet_widen_factor RESNET_WIDEN_FACTOR] [--wide_resnet_k WIDE_RESNET_K] [--resnext_cardinality RESNEXT_CARDINALITY] [--input_type INPUT_TYPE] [--manual_seed MANUAL_SEED] [--accimage] [--output_topk OUTPUT_TOPK] [--file_type FILE_TYPE] [--tensorboard] [--distributed] [--dist_url DIST_URL] [--world_size WORLD_SIZE] optional arguments: -h, --help show this help message and exit --root_path ROOT_PATH Root directory path --video_path VIDEO_PATH Directory path of videos --annotation_path ANNOTATION_PATH Annotation file path --result_path RESULT_PATH Result directory path --dataset DATASET Used dataset (activitynet | kinetics | ucf101 | hmdb51) --n_classes N_CLASSES Number of classes (activitynet: 200, kinetics: 400 or 600, ucf101: 101, hmdb51: 51) --n_pretrain_classes N_PRETRAIN_CLASSES Number of classes of pretraining task.When using --pretrain_path, this must be set. --pretrain_path PRETRAIN_PATH Pretrained model path (.pth). --ft_begin_module FT_BEGIN_MODULE Module name of beginning of fine-tuning(conv1, layer1, fc, denseblock1, classifier, ...).The default means all layers are fine-tuned. --sample_size SAMPLE_SIZE Height and width of inputs --sample_duration SAMPLE_DURATION Temporal duration of inputs --sample_t_stride SAMPLE_T_STRIDE If larger than 1, input frames are subsampled with the stride. --train_crop TRAIN_CROP Spatial cropping method in training. random is uniform. corner is selection from 4 corners and 1 center. (random | corner | center) --train_crop_min_scale TRAIN_CROP_MIN_SCALE Min scale for random cropping in training --train_crop_min_ratio TRAIN_CROP_MIN_RATIO Min aspect ratio for random cropping in training --no_hflip If true holizontal flipping is not performed. --colorjitter If true colorjitter is performed. --train_t_crop TRAIN_T_CROP Temporal cropping method in training. random is uniform. (random | center) --learning_rate LEARNING_RATE Initial learning rate(divided by 10 while training by lr scheduler) --momentum MOMENTUM Momentum --dampening DAMPENING dampening of SGD --weight_decay WEIGHT_DECAY Weight Decay --mean_dataset MEAN_DATASET dataset for mean values of mean subtraction(activitynet | kinetics | 0.5) --no_mean_norm If true, inputs are not normalized by mean. --no_std_norm If true, inputs are not normalized by standard deviation. --value_scale VALUE_SCALE If 1, range of inputs is [0-1]. If 255, range of inputs is [0-255]. --nesterov Nesterov momentum --optimizer OPTIMIZER Currently only support SGD --lr_scheduler LR_SCHEDULER Type of LR scheduler (multistep | plateau) --multistep_milestones MULTISTEP_MILESTONES [MULTISTEP_MILESTONES ...] Milestones of LR scheduler. See documentation of MultistepLR. --overwrite_milestones If true, overwriting multistep_milestones when resuming training. --plateau_patience PLATEAU_PATIENCE Patience of LR scheduler. See documentation of ReduceLROnPlateau. --batch_size BATCH_SIZE Batch Size --inference_batch_size INFERENCE_BATCH_SIZE Batch Size for inference. 0 means this is the same as batch_size. --batchnorm_sync If true, SyncBatchNorm is used instead of BatchNorm. --n_epochs N_EPOCHS Number of total epochs to run --n_val_samples N_VAL_SAMPLES Number of validation samples for each activity --resume_path RESUME_PATH Save data (.pth) of previous training --no_train If true, training is not performed. --no_val If true, validation is not performed. --inference If true, inference is performed. --inference_subset INFERENCE_SUBSET Used subset in inference (train | val | test) --inference_stride INFERENCE_STRIDE Stride of sliding window in inference. --inference_crop INFERENCE_CROP Cropping method in inference. (center | nocrop)When nocrop, fully convolutional inference is performed,and mini-batch consists of clips of one video. --inference_no_average If true, outputs for segments in a video are not averaged. --no_cuda If true, cuda is not used. --n_threads N_THREADS Number of threads for multi-thread loading --checkpoint CHECKPOINT Trained model is saved at every this epochs. --model MODEL (resnet | resnet2p1d | preresnet | wideresnet | resnext | densenet | --model_depth MODEL_DEPTH Depth of resnet (10 | 18 | 34 | 50 | 101) --conv1_t_size CONV1_T_SIZE Kernel size in t dim of conv1. --conv1_t_stride CONV1_T_STRIDE Stride in t dim of conv1. --no_max_pool If true, the max pooling after conv1 is removed. --resnet_shortcut RESNET_SHORTCUT Shortcut type of resnet (A | B) --resnet_widen_factor RESNET_WIDEN_FACTOR The number of feature maps of resnet is multiplied by this value --wide_resnet_k WIDE_RESNET_K Wide resnet k --resnext_cardinality RESNEXT_CARDINALITY ResNeXt cardinality --input_type INPUT_TYPE (rgb | flow) --manual_seed MANUAL_SEED Manually set random seed --accimage If true, accimage is used to load images. --output_topk OUTPUT_TOPK Top-k scores are saved in json file. --file_type FILE_TYPE (jpg | hdf5) --tensorboard If true, output tensorboard log file. --distributed Use multi-processing distributed training to launch N processes per node, which has N GPUs. --dist_url DIST_URL url used to set up distributed training --world_size WORLD_SIZE number of nodes for distributed training ``` ## 1.4 解压下载好的UCF-101视频数据集 命令说明：unzip -q 视频数据集压缩包所在目录 -d 解压的目标文件夹 ![image](pic/下载并解压UCF-101视频数据集.png) ## 1.5 视频转图片 命令格式：python -m util_scripts.generate_video_jpgs 视频地址 生成图片的保存地址 ucf101 将解压后的UCF-101视频文件转换为图片文件，保存在前面介绍的文件路径里面（~data/ucf_videos/jpg）。因为此处用的是UCF数据集，截图中的kinetics_videos应替换为ucf_videos（更改后会更规范一点，当时实现时没注意到这方面，后面实现的小伙伴可以留意留意）。 ![image](pic/avi2jpg1.png) ![image](pic/avi2jpg2.png) ![image](pic/avi2jpg3.png) ## 1.6 生成 json 格式的注释文件 python -m util_scripts.ucf101_json annotation_dir_path jpg_video_dir_path dst_json_path - annotation_dir_path：txt注释文件地址 - jpg_video_dir_path：avi转jpg后，jpg地址 - dst_json_path：json文件存储路径，默认按要求设置在data目录下 ## 1.7 开始本地Pytorch训练 ```bash python main.py --root_path /home/kevinpang/Documents/CodeLibrary/3D-ResNet/Learning-Spatio-Temporal-Features-with-3D-ResNets-PyTorch/data --video_path /home/kevinpang/Documents/CodeLibrary/3D-ResNet/Learning-Spatio-Temporal-Features-with-3D-ResNets-PyTorch/data/kinetics_videos/jpg --annotation_path /home/kevinpang/Documents/CodeLibrary/3D-ResNet/Learning-Spatio-Temporal-Features-with-3D-ResNets-PyTorch/data/ucf101_01.json --result_path /home/kevinpang/Documents/CodeLibrary/3D-ResNet/Learning-Spatio-Temporal-Features-with-3D-ResNets-PyTorch/data/results --dataset ucf101 --model resnet --model_depth 50 --n_classes 101 --batch_size 32 --n_threads 16 --checkpoint 1 ``` batch_size此处经测试最终本地选用32这个值，因为原先的128会直接导致显存溢出。checkpoint是设置保存训练模型的epoch间隔，此处是每个1个保存一个，仅作测试，实际训练需根据个人实际需求调节。 ![image](pic/测试pytorch代码1.png) ![image](pic/测试pytorch代码2.png) ![image](pic/测试pytorch代码3.png) ![image](pic/测试pytorch代码4.png) Ps:运行上述代码时，Pytorch代码时的显存占用，供参考。 ![image](pic/运行Pytorch代码时的内存占用.png) # 2 在AI STudio用paddlpaddle实现 ## 2.1 解压UCF-101视频数据集 ```python !unzip -q data/data48916/UCF-101.zip -d work/data/ucf_101/video ``` ## 2.2 实现视频转图片 ```python !python -m util_scripts.generate_video_jpgs work/data/ucf_101/video/UCF-101 work/data/ucf_101/jpg ucf101 ``` ## 2.3 根据split01划分数据集 ```python !unzip -q split01_ucf.zip -d work/data/ ``` ### 使用论文提供的脚本文件实现视频转图片 ```python !python -m util_scripts.ucf101_json work/data/split01_ucf work/data/ucf_101/jpg work/data/ ``` Traceback (most recent call last): File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/runpy.py", line 193, in _run_module_as_main "__main__", mod_spec) File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/runpy.py", line 85, in _run_code exec(code, run_globals) File "/home/aistudio/util_scripts/ucf101_json.py", line 94, in args.video_path, dst_json_path) File "/home/aistudio/util_scripts/ucf101_json.py", line 45, in convert_ucf101_csv_to_json train_database = convert_csv_to_dict(train_csv_path, 'training') File "/home/aistudio/util_scripts/ucf101_json.py", line 11, in convert_csv_to_dict data = pd.read_csv(csv_path, delimiter=' ', header=None) File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/pandas/io/parsers.py", line 678, in parser_f return _read(filepath_or_buffer, kwds) File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/pandas/io/parsers.py", line 440, in _read parser = TextFileReader(filepath_or_buffer, **kwds) File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/pandas/io/parsers.py", line 787, in __init__ self._make_engine(self.engine) File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/pandas/io/parsers.py", line 1014, in _make_engine self._engine = CParserWrapper(self.f, **self.options) File "/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/pandas/io/parsers.py", line 1708, in __init__ self._reader = parsers.TextReader(src, **kwds) File "pandas/_libs/parsers.pyx", line 384, in pandas._libs.parsers.TextReader.__cinit__ File "pandas/_libs/parsers.pyx", line 695, in pandas._libs.parsers.TextReader._setup_parser_source FileNotFoundError: File b'work/data/split01_ucf/trainlist02.txt' does not exist ### 使用视频分类实践里面的方法实现视频转图片(本论文复现最终使用的是此数据) ```python import os import numpy as np import cv2 video_src_src_path = 'work/data/ucf_101/video/UCF-101' video_src_TAG_path = 'work/data/ucf_101/video/UCF-101_JPG' label_name = os.listdir(video_src_src_path) label_dir = {} index = 0 for i in label_name: if i.startswith('.'): continue label_dir[i] = index index += 1 video_src_path = os.path.join(video_src_src_path, i) video_save_path = os.path.join(video_src_TAG_path, i) + '_jpg' # video_save_path = os.path.join(video_src_TAG_path, i) if not os.path.exists(video_src_TAG_path): os.mkdir(video_src_TAG_path) if not os.path.exists(video_save_path): os.mkdir(video_save_path) videos = os.listdir(video_src_path) # 过滤出avi文件 videos = filter(lambda x: x.endswith('avi'), videos) for each_video in videos: each_video_name, _ = each_video.split('.') if not os.path.exists(video_save_path + '/' + each_video_name): os.mkdir(video_save_path + '/' + each_video_name) each_video_save_full_path = os.path.join(video_save_path, each_video_name) + '/' each_video_full_path = os.path.join(video_src_path, each_video) cap = cv2.VideoCapture(each_video_full_path) frame_count = 1 success = True while success: success, frame = cap.read() # print('read a new frame:', success) params = [] params.append(1) if success: # cv2.imwrite(each_video_save_full_path + each_video_name + "_%d.jpg" % frame_count, frame, params) cv2.imwrite(each_video_save_full_path + each_video_name + "_%d.jpg" % frame_count, frame, [int(cv2.IMWRITE_JPEG_QUALITY), 75]) frame_count += 1 cap.release() np.save('label_dir.npy', label_dir) # print(label_dir) ``` ```python import os import numpy as np import cv2 import sys import glob import pickle from multiprocessing import Pool label_dic = np.load('label_dir.npy', allow_pickle=True).item() # print(label_dic) source_dir = 'work/data/ucf_101/video/UCF-101_JPG' # 图片保存路径 target_train_dir = 'work/data/ucf_101/video/UCF-101_JPG/train' # 训练集保存路径 target_test_dir = 'work/data/ucf_101/video/UCF-101_JPG/test' # 测试集保存路径 target_val_dir = 'work/data/ucf_101/video/UCF-101_JPG/val' # 验证集保存路径 if not os.path.exists(target_train_dir): os.mkdir(target_train_dir) if not os.path.exists(target_test_dir): os.mkdir(target_test_dir) if not os.path.exists(target_val_dir): os.mkdir(target_val_dir) for key in label_dic: each_mulu = key + '_jpg' # each_mulu = key print(each_mulu, key) label_dir = os.path.join(source_dir, each_mulu) label_mulu = os.listdir(label_dir) tag = 1 # print(label_dir) # print(len(label_mulu)) for each_label_mulu in label_mulu: image_file = os.listdir(os.path.join(label_dir, each_label_mulu)) image_file.sort() image_name = image_file[0][:-6] image_num = len(image_file) frame = [] vid = image_name # print(vid) sp = int(vid[-6:-4]) ####根据ucf101的split01，通过文件夹名字划分数据集 # print(sp) for i in range(image_num): image_path = os.path.join(os.path.join(label_dir, each_label_mulu), image_name + '_' + str(i+1) + '.jpg') frame.append(image_path) output_pkl = vid + '.pkl' if sp >= 8: output_pkl = os.path.join(target_train_dir, output_pkl) else: output_pkl = os.path.join(target_val_dir, output_pkl) tag += 1 f = open(output_pkl, 'wb') pickle.dump((vid, label_dic[key], frame), f, -1) f.close() ``` JumpRope_jpg JumpRope PoleVault_jpg PoleVault HorseRace_jpg HorseRace Skijet_jpg Skijet Nunchucks_jpg Nunchucks BoxingSpeedBag_jpg BoxingSpeedBag Knitting_jpg Knitting Bowling_jpg Bowling BenchPress_jpg BenchPress SumoWrestling_jpg SumoWrestling SoccerPenalty_jpg SoccerPenalty StillRings_jpg StillRings PlayingTabla_jpg PlayingTabla HulaHoop_jpg HulaHoop Punch_jpg Punch PullUps_jpg PullUps WallPushups_jpg WallPushups PlayingSitar_jpg PlayingSitar MilitaryParade_jpg MilitaryParade SalsaSpin_jpg SalsaSpin YoYo_jpg YoYo BreastStroke_jpg BreastStroke CuttingInKitchen_jpg CuttingInKitchen PlayingPiano_jpg PlayingPiano Haircut_jpg Haircut BlowingCandles_jpg BlowingCandles Diving_jpg Diving Basketball_jpg Basketball FloorGymnastics_jpg FloorGymnastics BandMarching_jpg BandMarching IceDancing_jpg IceDancing RockClimbingIndoor_jpg RockClimbingIndoor TaiChi_jpg TaiChi ShavingBeard_jpg ShavingBeard SkyDiving_jpg SkyDiving PlayingGuitar_jpg PlayingGuitar WalkingWithDog_jpg WalkingWithDog HandstandWalking_jpg HandstandWalking ParallelBars_jpg ParallelBars HandstandPushups_jpg HandstandPushups TableTennisShot_jpg TableTennisShot HeadMassage_jpg HeadMassage PommelHorse_jpg PommelHorse FieldHockeyPenalty_jpg FieldHockeyPenalty TrampolineJumping_jpg TrampolineJumping JugglingBalls_jpg JugglingBalls LongJump_jpg LongJump Surfing_jpg Surfing Hammering_jpg Hammering BoxingPunchingBag_jpg BoxingPunchingBag PlayingViolin_jpg PlayingViolin BalanceBeam_jpg BalanceBeam PlayingFlute_jpg PlayingFlute BasketballDunk_jpg BasketballDunk JavelinThrow_jpg JavelinThrow Rowing_jpg Rowing PizzaTossing_jpg PizzaTossing CliffDiving_jpg CliffDiving HighJump_jpg HighJump Mixing_jpg Mixing VolleyballSpiking_jpg VolleyballSpiking ApplyLipstick_jpg ApplyLipstick WritingOnBoard_jpg WritingOnBoard HammerThrow_jpg HammerThrow Drumming_jpg Drumming SkateBoarding_jpg SkateBoarding RopeClimbing_jpg RopeClimbing JumpingJack_jpg JumpingJack BaseballPitch_jpg BaseballPitch UnevenBars_jpg UnevenBars Rafting_jpg Rafting MoppingFloor_jpg MoppingFloor BabyCrawling_jpg BabyCrawling Biking_jpg Biking TennisSwing_jpg TennisSwing CleanAndJerk_jpg CleanAndJerk FrisbeeCatch_jpg FrisbeeCatch BrushingTeeth_jpg BrushingTeeth Typing_jpg Typing CricketBowling_jpg CricketBowling BodyWeightSquats_jpg BodyWeightSquats CricketShot_jpg CricketShot FrontCrawl_jpg FrontCrawl Shotput_jpg Shotput PlayingDhol_jpg PlayingDhol Billiards_jpg Billiards Kayaking_jpg Kayaking PlayingCello_jpg PlayingCello Archery_jpg Archery BlowDryHair_jpg BlowDryHair GolfSwing_jpg GolfSwing SoccerJuggling_jpg SoccerJuggling Swing_jpg Swing PushUps_jpg PushUps PlayingDaf_jpg PlayingDaf ApplyEyeMakeup_jpg ApplyEyeMakeup ThrowDiscus_jpg ThrowDiscus Skiing_jpg Skiing Lunges_jpg Lunges Fencing_jpg Fencing HorseRiding_jpg HorseRiding ```python import os data_dir = 'work/data/ucf_101/video/UCF-101_JPG/' train_data = os.listdir(data_dir + 'train') train_data = [x for x in train_data if not x.startswith('.')] print(len(train_data)) # test_data = os.listdir(data_dir + 'test') # test_data = [x for x in test_data if not x.startswith('.')] # print(len(test_data)) val_data = os.listdir(data_dir + 'val') val_data = [x for x in val_data if not x.startswith('.')] print(len(val_data)) f = open('work/data/ucf_101/video/UCF-101_JPG/train.list', 'w') for line in train_data: f.write(data_dir + 'train/' + line + '\n') f.close() # f = open('work/data/ucf_101/video/UCF-101_JPG/test.list', 'w') # for line in test_data: # f.write(data_dir + 'test/' + line + '\n') # f.close() f = open('work/data/ucf_101/video/UCF-101_JPG/val.list', 'w') for line in val_data: f.write(data_dir + 'val/' + line + '\n') f.close() ``` 9537 3783 ## 2.4 训练 ```python !python train.py --use_gpu True --epoch 50 --batch_size 64 --save_dir 'checkpoints_models' --config 'configs/res3d_L50.txt' --pretrain True ``` {'MODEL': {'name': 'Res3D', 'format': 'pkl', 'num_classes': 101, 'seg_num': 1, 'seglen': 8, 'image_mean': [0.485, 0.456, 0.406], 'image_std': [0.229, 0.224, 0.225], 'num_layers': 50}, 'TRAIN': {'epoch': 45, 'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 36, 'use_gpu': True, 'num_gpus': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/train.list', 'learning_rate': 0.01, 'learning_rate_decay': 0.1, 'l2_weight_decay': 0.0001, 'momentum': 0.9, 'total_videos': 80}, 'VALID': {'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/val.list'}, 'TEST': {'seg_num': 1, 'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/test.list'}, 'INFER': {'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/test.list'}} W0905 22:04:32.097486 29064 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 10.1, Runtime API Version: 9.0 W0905 22:04:32.101613 29064 device_context.cc:260] device: 0, cuDNN Version: 7.3. Loss at epoch 0 step 0: [14.85315], acc: [0.] Loss at epoch 0 step 12: [6.6796126], acc: [0.0625] Loss at epoch 0 step 24: [5.414842], acc: [0.078125] Loss at epoch 0 step 36: [4.329979], acc: [0.109375] Loss at epoch 0 step 48: [4.1428776], acc: [0.15625] Loss at epoch 0 step 60: [3.3291516], acc: [0.296875] Loss at epoch 0 step 72: [3.5765717], acc: [0.1875] Loss at epoch 0 step 84: [3.2602062], acc: [0.28125] Loss at epoch 0 step 96: [2.7732463], acc: [0.375] Loss at epoch 0 step 108: [2.5205185], acc: [0.390625] Loss at epoch 0 step 120: [2.3515425], acc: [0.375] Loss at epoch 0 step 132: [2.3940594], acc: [0.390625] Loss at epoch 0 step 144: [2.6096253], acc: [0.328125] Loss at epoch 1 step 0: [2.2513657], acc: [0.484375] Loss at epoch 1 step 12: [1.9614946], acc: [0.515625] Loss at epoch 1 step 24: [1.814927], acc: [0.609375] Loss at epoch 1 step 36: [1.6050638], acc: [0.59375] Loss at epoch 1 step 48: [1.8906935], acc: [0.546875] Loss at epoch 1 step 60: [1.5568355], acc: [0.609375] Loss at epoch 1 step 72: [1.654619], acc: [0.578125] Loss at epoch 1 step 84: [2.1530056], acc: [0.5] Loss at epoch 1 step 96: [1.3266186], acc: [0.640625] Loss at epoch 1 step 108: [1.5023375], acc: [0.59375] Loss at epoch 1 step 120: [1.3825148], acc: [0.65625] Loss at epoch 1 step 132: [0.9803108], acc: [0.734375] Loss at epoch 1 step 144: [1.4183452], acc: [0.609375] Loss at epoch 2 step 0: [1.363558], acc: [0.609375] Loss at epoch 2 step 12: [1.3321362], acc: [0.65625] Loss at epoch 2 step 24: [1.1997626], acc: [0.640625] Loss at epoch 2 step 36: [1.3337362], acc: [0.5625] Loss at epoch 2 step 48: [1.2206215], acc: [0.703125] Loss at epoch 2 step 60: [1.0951669], acc: [0.75] Loss at epoch 2 step 72: [1.0150272], acc: [0.75] Loss at epoch 2 step 84: [0.59039515], acc: [0.875] Loss at epoch 2 step 96: [0.98881704], acc: [0.765625] Loss at epoch 2 step 108: [0.8799726], acc: [0.8125] Loss at epoch 2 step 120: [1.0080926], acc: [0.765625] Loss at epoch 2 step 132: [0.80505145], acc: [0.875] Loss at epoch 2 step 144: [0.69941515], acc: [0.796875] Loss at epoch 3 step 0: [0.8660723], acc: [0.828125] Loss at epoch 3 step 12: [0.52552694], acc: [0.875] Loss at epoch 3 step 24: [0.7111227], acc: [0.765625] Loss at epoch 3 step 36: [0.76069826], acc: [0.796875] Loss at epoch 3 step 48: [0.64326614], acc: [0.828125] Loss at epoch 3 step 60: [0.69727623], acc: [0.84375] Loss at epoch 3 step 72: [0.9416874], acc: [0.75] Loss at epoch 3 step 84: [0.80490947], acc: [0.84375] Loss at epoch 3 step 96: [0.79475594], acc: [0.8125] Loss at epoch 3 step 108: [0.6326566], acc: [0.8125] Loss at epoch 3 step 120: [0.6236204], acc: [0.796875] Loss at epoch 3 step 132: [0.68786544], acc: [0.796875] Loss at epoch 3 step 144: [0.6987828], acc: [0.84375] Loss at epoch 4 step 0: [0.6295227], acc: [0.8125] Loss at epoch 4 step 12: [0.6604608], acc: [0.765625] Loss at epoch 4 step 24: [0.43752924], acc: [0.90625] Loss at epoch 4 step 36: [0.29913998], acc: [0.921875] Loss at epoch 4 step 48: [0.39409184], acc: [0.90625] Loss at epoch 4 step 60: [0.67640555], acc: [0.875] Loss at epoch 4 step 72: [0.4235146], acc: [0.875] Loss at epoch 4 step 84: [0.5107883], acc: [0.875] Loss at epoch 4 step 96: [0.5095406], acc: [0.8125] Loss at epoch 4 step 108: [0.5783402], acc: [0.8125] Loss at epoch 4 step 120: [0.4810606], acc: [0.90625] Loss at epoch 4 step 132: [0.70278645], acc: [0.84375] Loss at epoch 4 step 144: [0.5491826], acc: [0.84375] Loss at epoch 5 step 0: [0.44033235], acc: [0.90625] Loss at epoch 5 step 12: [0.37375376], acc: [0.90625] Loss at epoch 5 step 24: [0.402488], acc: [0.921875] Loss at epoch 5 step 36: [0.41290492], acc: [0.921875] Loss at epoch 5 step 48: [0.33651885], acc: [0.9375] Loss at epoch 5 step 60: [0.40448695], acc: [0.921875] Loss at epoch 5 step 72: [0.3146685], acc: [0.90625] Loss at epoch 5 step 84: [0.48406652], acc: [0.859375] Loss at epoch 5 step 96: [0.36037546], acc: [0.921875] Loss at epoch 5 step 108: [0.4160246], acc: [0.890625] Loss at epoch 5 step 120: [0.29046], acc: [0.9375] Loss at epoch 5 step 132: [0.3818245], acc: [0.890625] Loss at epoch 5 step 144: [0.41643608], acc: [0.90625] Loss at epoch 6 step 0: [0.47544712], acc: [0.875] Loss at epoch 6 step 12: [0.4636262], acc: [0.84375] Loss at epoch 6 step 24: [0.3017388], acc: [0.921875] Loss at epoch 6 step 36: [0.27119005], acc: [0.921875] Loss at epoch 6 step 48: [0.38429826], acc: [0.890625] Loss at epoch 6 step 60: [0.39860123], acc: [0.890625] Loss at epoch 6 step 72: [0.26195326], acc: [0.890625] Loss at epoch 6 step 84: [0.24620596], acc: [0.9375] Loss at epoch 6 step 96: [0.367297], acc: [0.890625] Loss at epoch 6 step 108: [0.28543663], acc: [0.9375] Loss at epoch 6 step 120: [0.37498397], acc: [0.90625] Loss at epoch 6 step 132: [0.24652094], acc: [0.921875] Loss at epoch 6 step 144: [0.31677723], acc: [0.9375] Loss at epoch 7 step 0: [0.1496607], acc: [0.984375] Loss at epoch 7 step 12: [0.30106866], acc: [0.9375] Loss at epoch 7 step 24: [0.2952159], acc: [0.921875] Loss at epoch 7 step 36: [0.16807702], acc: [0.984375] Loss at epoch 7 step 48: [0.27179495], acc: [0.921875] Loss at epoch 7 step 60: [0.11121802], acc: [1.] Loss at epoch 7 step 72: [0.1859386], acc: [0.96875] Loss at epoch 7 step 84: [0.16966134], acc: [0.96875] Loss at epoch 7 step 96: [0.15023518], acc: [0.96875] Loss at epoch 7 step 108: [0.22545943], acc: [0.953125] Loss at epoch 7 step 120: [0.21359515], acc: [0.953125] Loss at epoch 7 step 132: [0.22624011], acc: [0.9375] Loss at epoch 7 step 144: [0.28296125], acc: [0.9375] Loss at epoch 8 step 0: [0.2438523], acc: [0.953125] Loss at epoch 8 step 12: [0.20319524], acc: [0.96875] Loss at epoch 8 step 24: [0.08904399], acc: [1.] Loss at epoch 8 step 36: [0.24717543], acc: [0.9375] Loss at epoch 8 step 48: [0.1869955], acc: [0.9375] Loss at epoch 8 step 60: [0.27859107], acc: [0.921875] Loss at epoch 8 step 72: [0.15005513], acc: [0.984375] Loss at epoch 8 step 84: [0.3568893], acc: [0.90625] Loss at epoch 8 step 96: [0.24407665], acc: [0.9375] Loss at epoch 8 step 108: [0.1589992], acc: [0.96875] Loss at epoch 8 step 120: [0.3000587], acc: [0.921875] Loss at epoch 8 step 132: [0.15202102], acc: [0.984375] Loss at epoch 8 step 144: [0.17012021], acc: [0.96875] Loss at epoch 9 step 0: [0.16879855], acc: [0.953125] Loss at epoch 9 step 12: [0.19536692], acc: [0.953125] Loss at epoch 9 step 24: [0.10351394], acc: [1.] Loss at epoch 9 step 36: [0.11839722], acc: [0.96875] Loss at epoch 9 step 48: [0.22573878], acc: [0.96875] Loss at epoch 9 step 60: [0.11900575], acc: [0.984375] Loss at epoch 9 step 72: [0.18017237], acc: [0.953125] Loss at epoch 9 step 84: [0.10396101], acc: [0.984375] Loss at epoch 9 step 96: [0.17987195], acc: [0.953125] Loss at epoch 9 step 108: [0.16871275], acc: [0.9375] Loss at epoch 9 step 120: [0.16428079], acc: [0.96875] Loss at epoch 9 step 132: [0.21758851], acc: [0.953125] Loss at epoch 9 step 144: [0.15577325], acc: [0.96875] Loss at epoch 10 step 0: [0.2524661], acc: [0.9375] Loss at epoch 10 step 12: [0.29477394], acc: [0.90625] Loss at epoch 10 step 24: [0.16717619], acc: [0.984375] Loss at epoch 10 step 36: [0.10647671], acc: [0.984375] Loss at epoch 10 step 48: [0.17209145], acc: [0.96875] Loss at epoch 10 step 60: [0.19037256], acc: [0.921875] Loss at epoch 10 step 72: [0.13078612], acc: [0.984375] Loss at epoch 10 step 84: [0.21022335], acc: [0.96875] Loss at epoch 10 step 96: [0.22082496], acc: [0.9375] Loss at epoch 10 step 108: [0.2115297], acc: [0.96875] Loss at epoch 10 step 120: [0.18263528], acc: [0.9375] Loss at epoch 10 step 132: [0.208135], acc: [0.953125] Loss at epoch 10 step 144: [0.09450705], acc: [0.984375] Loss at epoch 11 step 0: [0.0934329], acc: [0.984375] Loss at epoch 11 step 12: [0.10507298], acc: [0.984375] Loss at epoch 11 step 24: [0.09788331], acc: [0.96875] Loss at epoch 11 step 36: [0.09634262], acc: [0.96875] Loss at epoch 11 step 48: [0.1056648], acc: [0.96875] Loss at epoch 11 step 60: [0.12958457], acc: [0.953125] Loss at epoch 11 step 72: [0.09875301], acc: [0.984375] Loss at epoch 11 step 84: [0.12760097], acc: [1.] Loss at epoch 11 step 96: [0.1525197], acc: [0.96875] Loss at epoch 11 step 108: [0.29978272], acc: [0.921875] Loss at epoch 11 step 120: [0.17637053], acc: [0.96875] Loss at epoch 11 step 132: [0.05010311], acc: [1.] Loss at epoch 11 step 144: [0.16990344], acc: [0.96875] Loss at epoch 12 step 0: [0.09047768], acc: [0.984375] Loss at epoch 12 step 12: [0.11322624], acc: [0.984375] Loss at epoch 12 step 24: [0.13600557], acc: [0.984375] Loss at epoch 12 step 36: [0.16450971], acc: [0.9375] Loss at epoch 12 step 48: [0.04274707], acc: [1.] Loss at epoch 12 step 60: [0.12501934], acc: [0.953125] Loss at epoch 12 step 72: [0.2243658], acc: [0.921875] Loss at epoch 12 step 84: [0.12063926], acc: [0.953125] Loss at epoch 12 step 96: [0.12818214], acc: [0.96875] Loss at epoch 12 step 108: [0.17364393], acc: [0.9375] Loss at epoch 12 step 120: [0.2114651], acc: [0.9375] Loss at epoch 12 step 132: [0.1331478], acc: [0.96875] Loss at epoch 12 step 144: [0.12901652], acc: [0.984375] Loss at epoch 13 step 0: [0.12000009], acc: [0.96875] Loss at epoch 13 step 12: [0.13576351], acc: [0.96875] Loss at epoch 13 step 24: [0.07339023], acc: [1.] Loss at epoch 13 step 36: [0.15410402], acc: [0.953125] Loss at epoch 13 step 48: [0.0427796], acc: [1.] Loss at epoch 13 step 60: [0.28730157], acc: [0.9375] Loss at epoch 13 step 72: [0.05982348], acc: [1.] Loss at epoch 13 step 84: [0.14123602], acc: [0.984375] Loss at epoch 13 step 96: [0.09428114], acc: [0.984375] Loss at epoch 13 step 108: [0.13396299], acc: [0.953125] Loss at epoch 13 step 120: [0.21153726], acc: [0.9375] Loss at epoch 13 step 132: [0.08319372], acc: [1.] Loss at epoch 13 step 144: [0.10827542], acc: [0.984375] Loss at epoch 14 step 0: [0.12854688], acc: [0.96875] Loss at epoch 14 step 12: [0.03872731], acc: [1.] Loss at epoch 14 step 24: [0.14342298], acc: [0.96875] Loss at epoch 14 step 36: [0.11444628], acc: [0.984375] Loss at epoch 14 step 48: [0.07736364], acc: [0.96875] Loss at epoch 14 step 60: [0.03748611], acc: [1.] Loss at epoch 14 step 72: [0.07666603], acc: [1.] Loss at epoch 14 step 84: [0.08613737], acc: [0.984375] Loss at epoch 14 step 96: [0.09753634], acc: [0.96875] Loss at epoch 14 step 108: [0.06769439], acc: [0.984375] Loss at epoch 14 step 120: [0.22970295], acc: [0.953125] Loss at epoch 14 step 132: [0.08329549], acc: [0.984375] Loss at epoch 14 step 144: [0.0625074], acc: [1.] Loss at epoch 15 step 0: [0.2300813], acc: [0.953125] Loss at epoch 15 step 12: [0.13319677], acc: [0.96875] Loss at epoch 15 step 24: [0.1139845], acc: [0.953125] Loss at epoch 15 step 36: [0.10999233], acc: [0.953125] Loss at epoch 15 step 48: [0.1930984], acc: [0.921875] Loss at epoch 15 step 60: [0.14966239], acc: [0.953125] Loss at epoch 15 step 72: [0.05806785], acc: [1.] Loss at epoch 15 step 84: [0.13039914], acc: [0.984375] Loss at epoch 15 step 96: [0.08497517], acc: [0.96875] Loss at epoch 15 step 108: [0.05144798], acc: [1.] Loss at epoch 15 step 120: [0.11188055], acc: [0.96875] Loss at epoch 15 step 132: [0.0436349], acc: [1.] Loss at epoch 15 step 144: [0.07105402], acc: [0.984375] Loss at epoch 16 step 0: [0.05386063], acc: [1.] Loss at epoch 16 step 12: [0.08448453], acc: [0.984375] Loss at epoch 16 step 24: [0.05688261], acc: [0.984375] Loss at epoch 16 step 36: [0.07019769], acc: [1.] Loss at epoch 16 step 48: [0.09562925], acc: [0.984375] Loss at epoch 16 step 60: [0.07893139], acc: [0.984375] Loss at epoch 16 step 72: [0.12622833], acc: [0.96875] Loss at epoch 16 step 84: [0.22259906], acc: [0.96875] Loss at epoch 16 step 96: [0.08087094], acc: [1.] Loss at epoch 16 step 108: [0.19678888], acc: [0.953125] Loss at epoch 16 step 120: [0.09612665], acc: [0.96875] Loss at epoch 16 step 132: [0.07336933], acc: [0.984375] Loss at epoch 16 step 144: [0.273827], acc: [0.921875] Loss at epoch 17 step 0: [0.03031229], acc: [1.] Loss at epoch 17 step 12: [0.05119709], acc: [0.984375] Loss at epoch 17 step 24: [0.12925656], acc: [0.984375] Loss at epoch 17 step 36: [0.03025132], acc: [1.] Loss at epoch 17 step 48: [0.08053686], acc: [0.984375] Loss at epoch 17 step 60: [0.13075587], acc: [0.953125] Loss at epoch 17 step 72: [0.11190623], acc: [0.984375] Loss at epoch 17 step 84: [0.11468533], acc: [0.984375] Loss at epoch 17 step 96: [0.09129784], acc: [0.984375] Loss at epoch 17 step 108: [0.0594643], acc: [1.] Loss at epoch 17 step 120: [0.13537651], acc: [0.96875] Loss at epoch 17 step 132: [0.04839276], acc: [1.] Loss at epoch 17 step 144: [0.05619723], acc: [0.984375] Loss at epoch 18 step 0: [0.09760777], acc: [0.984375] Loss at epoch 18 step 12: [0.0785483], acc: [0.984375] Loss at epoch 18 step 24: [0.10290235], acc: [0.953125] Loss at epoch 18 step 36: [0.06441458], acc: [0.984375] Loss at epoch 18 step 48: [0.11317109], acc: [0.96875] Loss at epoch 18 step 60: [0.13344187], acc: [0.984375] Loss at epoch 18 step 72: [0.09341757], acc: [0.984375] Loss at epoch 18 step 84: [0.02922366], acc: [1.] Loss at epoch 18 step 96: [0.13897681], acc: [0.96875] Loss at epoch 18 step 108: [0.13234375], acc: [0.984375] Loss at epoch 18 step 120: [0.1212811], acc: [0.9375] Loss at epoch 18 step 132: [0.04489965], acc: [1.] Loss at epoch 18 step 144: [0.1560329], acc: [0.96875] Loss at epoch 19 step 0: [0.14172162], acc: [0.984375] Loss at epoch 19 step 12: [0.02631309], acc: [1.] Loss at epoch 19 step 24: [0.0556977], acc: [0.984375] Loss at epoch 19 step 36: [0.05195104], acc: [1.] Loss at epoch 19 step 48: [0.02699609], acc: [1.] Loss at epoch 19 step 60: [0.14031532], acc: [0.96875] Loss at epoch 19 step 72: [0.04408358], acc: [1.] Loss at epoch 19 step 84: [0.09961557], acc: [0.953125] Loss at epoch 19 step 96: [0.03455441], acc: [0.984375] Loss at epoch 19 step 108: [0.09078674], acc: [0.984375] Loss at epoch 19 step 120: [0.1229469], acc: [0.96875] Loss at epoch 19 step 132: [0.11646602], acc: [0.984375] Loss at epoch 19 step 144: [0.04109331], acc: [1.] Loss at epoch 20 step 0: [0.04554827], acc: [0.984375] Loss at epoch 20 step 12: [0.03125779], acc: [1.] Loss at epoch 20 step 24: [0.13076854], acc: [0.984375] Loss at epoch 20 step 36: [0.08232777], acc: [0.984375] Loss at epoch 20 step 48: [0.04520696], acc: [1.] Loss at epoch 20 step 60: [0.02101058], acc: [1.] Loss at epoch 20 step 72: [0.07459157], acc: [0.984375] Loss at epoch 20 step 84: [0.06713352], acc: [0.96875] Loss at epoch 20 step 96: [0.0294167], acc: [1.] Loss at epoch 20 step 108: [0.05508721], acc: [0.984375] Loss at epoch 20 step 120: [0.11334416], acc: [0.96875] Loss at epoch 20 step 132: [0.0358643], acc: [1.] Loss at epoch 20 step 144: [0.07875131], acc: [0.984375] Loss at epoch 21 step 0: [0.13761021], acc: [0.984375] Loss at epoch 21 step 12: [0.05790991], acc: [0.984375] Loss at epoch 21 step 24: [0.058786], acc: [0.984375] Loss at epoch 21 step 36: [0.07185003], acc: [0.984375] Loss at epoch 21 step 48: [0.0523248], acc: [0.984375] Loss at epoch 21 step 60: [0.07980347], acc: [0.984375] Loss at epoch 21 step 72: [0.10166893], acc: [0.96875] Loss at epoch 21 step 84: [0.03680546], acc: [0.984375] Loss at epoch 21 step 96: [0.05135682], acc: [1.] Loss at epoch 21 step 108: [0.06743972], acc: [0.984375] Loss at epoch 21 step 120: [0.02514549], acc: [1.] Loss at epoch 21 step 132: [0.02343409], acc: [1.] Loss at epoch 21 step 144: [0.06429012], acc: [0.984375] Loss at epoch 22 step 0: [0.07111311], acc: [0.984375] Loss at epoch 22 step 12: [0.03257249], acc: [1.] Loss at epoch 22 step 24: [0.13606302], acc: [0.96875] Loss at epoch 22 step 36: [0.02992693], acc: [1.] Loss at epoch 22 step 48: [0.05402429], acc: [0.984375] Loss at epoch 22 step 60: [0.16036783], acc: [0.96875] Loss at epoch 22 step 72: [0.10111773], acc: [0.96875] Loss at epoch 22 step 84: [0.04418909], acc: [1.] Loss at epoch 22 step 96: [0.08994643], acc: [0.984375] Loss at epoch 22 step 108: [0.03462774], acc: [1.] Loss at epoch 22 step 120: [0.07127047], acc: [0.984375] Loss at epoch 22 step 132: [0.06169221], acc: [0.984375] Loss at epoch 22 step 144: [0.03054155], acc: [1.] Loss at epoch 23 step 0: [0.02151599], acc: [1.] Loss at epoch 23 step 12: [0.13141753], acc: [0.96875] Loss at epoch 23 step 24: [0.02565545], acc: [1.] Loss at epoch 23 step 36: [0.0405319], acc: [1.] Loss at epoch 23 step 48: [0.03404613], acc: [1.] Loss at epoch 23 step 60: [0.02181065], acc: [1.] Loss at epoch 23 step 72: [0.07211412], acc: [0.96875] Loss at epoch 23 step 84: [0.05584048], acc: [1.] Loss at epoch 23 step 96: [0.09588777], acc: [0.984375] Loss at epoch 23 step 108: [0.03304356], acc: [1.] Loss at epoch 23 step 120: [0.07091898], acc: [0.984375] Loss at epoch 23 step 132: [0.00978923], acc: [1.] Loss at epoch 23 step 144: [0.03111069], acc: [1.] Loss at epoch 24 step 0: [0.07336287], acc: [0.984375] Loss at epoch 24 step 12: [0.03198887], acc: [1.] Loss at epoch 24 step 24: [0.03469952], acc: [0.984375] Loss at epoch 24 step 36: [0.01714237], acc: [1.] Loss at epoch 24 step 48: [0.01813993], acc: [1.] Loss at epoch 24 step 60: [0.0522625], acc: [1.] Loss at epoch 24 step 72: [0.05092136], acc: [0.984375] Loss at epoch 24 step 84: [0.02400209], acc: [1.] Loss at epoch 24 step 96: [0.10486159], acc: [0.953125] Loss at epoch 24 step 108: [0.05936999], acc: [0.984375] Loss at epoch 24 step 120: [0.07598761], acc: [0.984375] Loss at epoch 24 step 132: [0.11239941], acc: [0.984375] Loss at epoch 24 step 144: [0.14368948], acc: [0.9375] Loss at epoch 25 step 0: [0.07386638], acc: [0.984375] Loss at epoch 25 step 12: [0.06467087], acc: [0.984375] Loss at epoch 25 step 24: [0.06013472], acc: [1.] Loss at epoch 25 step 36: [0.04554974], acc: [1.] Loss at epoch 25 step 48: [0.03613939], acc: [0.984375] Loss at epoch 25 step 60: [0.02343844], acc: [1.] Loss at epoch 25 step 72: [0.05930378], acc: [1.] Loss at epoch 25 step 84: [0.03189167], acc: [0.984375] Loss at epoch 25 step 96: [0.02557785], acc: [1.] Loss at epoch 25 step 108: [0.04231068], acc: [1.] Loss at epoch 25 step 120: [0.05833966], acc: [0.984375] Loss at epoch 25 step 132: [0.06038568], acc: [0.984375] Loss at epoch 25 step 144: [0.01606149], acc: [1.] Loss at epoch 26 step 0: [0.03596912], acc: [1.] Loss at epoch 26 step 12: [0.12398455], acc: [0.96875] Loss at epoch 26 step 24: [0.01009746], acc: [1.] Loss at epoch 26 step 36: [0.06498827], acc: [0.984375] Loss at epoch 26 step 48: [0.01815913], acc: [1.] Loss at epoch 26 step 60: [0.03100191], acc: [0.984375] Loss at epoch 26 step 72: [0.03038638], acc: [1.] Loss at epoch 26 step 84: [0.06207973], acc: [0.984375] Loss at epoch 26 step 96: [0.05353945], acc: [0.984375] Loss at epoch 26 step 108: [0.02559943], acc: [0.984375] Loss at epoch 26 step 120: [0.03326371], acc: [0.984375] Loss at epoch 26 step 132: [0.0735067], acc: [0.96875] Loss at epoch 26 step 144: [0.06352422], acc: [0.984375] Loss at epoch 27 step 0: [0.04096693], acc: [1.] Loss at epoch 27 step 12: [0.04134082], acc: [1.] Loss at epoch 27 step 24: [0.03708979], acc: [0.984375] Loss at epoch 27 step 36: [0.04432627], acc: [0.984375] Loss at epoch 27 step 48: [0.01704144], acc: [1.] Loss at epoch 27 step 60: [0.05820725], acc: [0.984375] Loss at epoch 27 step 72: [0.0134847], acc: [1.] Loss at epoch 27 step 84: [0.0256593], acc: [0.984375] Loss at epoch 27 step 96: [0.02774268], acc: [1.] Loss at epoch 27 step 108: [0.08903482], acc: [0.984375] Loss at epoch 27 step 120: [0.01867808], acc: [1.] Loss at epoch 27 step 132: [0.05625074], acc: [0.984375] Loss at epoch 27 step 144: [0.04535157], acc: [1.] Loss at epoch 28 step 0: [0.01853942], acc: [1.] Loss at epoch 28 step 12: [0.05177148], acc: [1.] Loss at epoch 28 step 24: [0.06037938], acc: [0.984375] Loss at epoch 28 step 36: [0.02143999], acc: [1.] Loss at epoch 28 step 48: [0.01737234], acc: [1.] Loss at epoch 28 step 60: [0.02225753], acc: [1.] Loss at epoch 28 step 72: [0.02325612], acc: [1.] Loss at epoch 28 step 84: [0.06157194], acc: [0.96875] Loss at epoch 28 step 96: [0.07357164], acc: [0.984375] Loss at epoch 28 step 108: [0.09568638], acc: [0.984375] Loss at epoch 28 step 120: [0.02652951], acc: [1.] Loss at epoch 28 step 132: [0.04191494], acc: [1.] Loss at epoch 28 step 144: [0.02180932], acc: [1.] Loss at epoch 29 step 0: [0.0146846], acc: [1.] Loss at epoch 29 step 12: [0.01233242], acc: [1.] Loss at epoch 29 step 24: [0.02196808], acc: [1.] Loss at epoch 29 step 36: [0.02617442], acc: [1.] Loss at epoch 29 step 48: [0.01640628], acc: [1.] Loss at epoch 29 step 60: [0.06456419], acc: [0.984375] Loss at epoch 29 step 72: [0.01993908], acc: [1.] Loss at epoch 29 step 84: [0.05435857], acc: [0.984375] Loss at epoch 29 step 96: [0.01356348], acc: [1.] Loss at epoch 29 step 108: [0.02805275], acc: [1.] Loss at epoch 29 step 120: [0.01442439], acc: [1.] Loss at epoch 29 step 132: [0.00675201], acc: [1.] Loss at epoch 29 step 144: [0.0628389], acc: [0.984375] Loss at epoch 30 step 0: [0.0230991], acc: [1.] Loss at epoch 30 step 12: [0.02196022], acc: [1.] Loss at epoch 30 step 24: [0.04323066], acc: [0.984375] Loss at epoch 30 step 36: [0.03947566], acc: [0.984375] Loss at epoch 30 step 48: [0.05507033], acc: [0.984375] Loss at epoch 30 step 60: [0.02097051], acc: [1.] Loss at epoch 30 step 72: [0.06954969], acc: [0.96875] Loss at epoch 30 step 84: [0.05194077], acc: [0.984375] Loss at epoch 30 step 96: [0.3182054], acc: [0.90625] Loss at epoch 30 step 108: [0.05387548], acc: [0.984375] Loss at epoch 30 step 120: [0.03082899], acc: [0.984375] Loss at epoch 30 step 132: [0.02838017], acc: [1.] Loss at epoch 30 step 144: [0.0786593], acc: [0.984375] Loss at epoch 31 step 0: [0.01530119], acc: [1.] Loss at epoch 31 step 12: [0.03183078], acc: [1.] Loss at epoch 31 step 24: [0.05290516], acc: [0.984375] Loss at epoch 31 step 36: [0.02267444], acc: [1.] Loss at epoch 31 step 48: [0.04164417], acc: [0.984375] Loss at epoch 31 step 60: [0.03736061], acc: [0.984375] Loss at epoch 31 step 72: [0.04186511], acc: [0.984375] Loss at epoch 31 step 84: [0.0226022], acc: [1.] Loss at epoch 31 step 96: [0.03843619], acc: [0.984375] Loss at epoch 31 step 108: [0.03947717], acc: [0.984375] Loss at epoch 31 step 120: [0.03992875], acc: [0.984375] Loss at epoch 31 step 132: [0.02740411], acc: [1.] Loss at epoch 31 step 144: [0.0499729], acc: [0.984375] Loss at epoch 32 step 0: [0.00776515], acc: [1.] Loss at epoch 32 step 12: [0.03891348], acc: [1.] Loss at epoch 32 step 24: [0.10477659], acc: [0.984375] Loss at epoch 32 step 36: [0.01969251], acc: [1.] Loss at epoch 32 step 48: [0.02965295], acc: [1.] Loss at epoch 32 step 60: [0.01365479], acc: [1.] Loss at epoch 32 step 72: [0.03200981], acc: [0.984375] Loss at epoch 32 step 84: [0.02956506], acc: [1.] Loss at epoch 32 step 96: [0.02138363], acc: [1.] Loss at epoch 32 step 108: [0.0884956], acc: [0.984375] Loss at epoch 32 step 120: [0.03419771], acc: [1.] Loss at epoch 32 step 132: [0.04211798], acc: [0.984375] Loss at epoch 32 step 144: [0.02565004], acc: [1.] Loss at epoch 33 step 0: [0.04761265], acc: [0.984375] Loss at epoch 33 step 12: [0.02598224], acc: [1.] Loss at epoch 33 step 24: [0.05979962], acc: [0.984375] Loss at epoch 33 step 36: [0.02802056], acc: [1.] Loss at epoch 33 step 48: [0.04433171], acc: [0.984375] Loss at epoch 33 step 60: [0.02380951], acc: [1.] Loss at epoch 33 step 72: [0.03001182], acc: [1.] Loss at epoch 33 step 84: [0.01514533], acc: [1.] Loss at epoch 33 step 96: [0.07400075], acc: [0.984375] Loss at epoch 33 step 108: [0.13431], acc: [0.96875] Loss at epoch 33 step 120: [0.09481361], acc: [0.96875] Loss at epoch 33 step 132: [0.02092645], acc: [1.] Loss at epoch 33 step 144: [0.02946315], acc: [1.] Loss at epoch 34 step 0: [0.02682259], acc: [1.] Loss at epoch 34 step 12: [0.03863063], acc: [0.984375] Loss at epoch 34 step 24: [0.11434168], acc: [0.96875] Loss at epoch 34 step 36: [0.0478663], acc: [0.984375] Loss at epoch 34 step 48: [0.04993972], acc: [0.984375] Loss at epoch 34 step 60: [0.01280891], acc: [1.] Loss at epoch 34 step 72: [0.02723259], acc: [1.] Loss at epoch 34 step 84: [0.04503974], acc: [0.984375] Loss at epoch 34 step 96: [0.05953965], acc: [0.984375] Loss at epoch 34 step 108: [0.02233556], acc: [1.] Loss at epoch 34 step 120: [0.02156727], acc: [1.] Loss at epoch 34 step 132: [0.06009539], acc: [0.96875] Loss at epoch 34 step 144: [0.02646555], acc: [1.] Loss at epoch 35 step 0: [0.01192369], acc: [1.] Loss at epoch 35 step 12: [0.01927817], acc: [1.] Loss at epoch 35 step 24: [0.03226478], acc: [0.984375] Loss at epoch 35 step 36: [0.01435088], acc: [1.] Loss at epoch 35 step 48: [0.03618512], acc: [0.984375] Loss at epoch 35 step 60: [0.02448768], acc: [0.984375] Loss at epoch 35 step 72: [0.01529774], acc: [1.] Loss at epoch 35 step 84: [0.01098083], acc: [1.] Loss at epoch 35 step 96: [0.01899394], acc: [1.] Loss at epoch 35 step 108: [0.04528006], acc: [1.] Loss at epoch 35 step 120: [0.06730434], acc: [0.96875] Loss at epoch 35 step 132: [0.04443285], acc: [0.984375] Loss at epoch 35 step 144: [0.04580284], acc: [0.984375] Loss at epoch 36 step 0: [0.01955256], acc: [1.] Loss at epoch 36 step 12: [0.03178887], acc: [1.] Loss at epoch 36 step 24: [0.08350191], acc: [0.984375] Loss at epoch 36 step 36: [0.0360631], acc: [0.984375] Loss at epoch 36 step 48: [0.05383827], acc: [0.984375] Loss at epoch 36 step 60: [0.02050314], acc: [1.] Loss at epoch 36 step 72: [0.04018402], acc: [1.] Loss at epoch 36 step 84: [0.0231873], acc: [1.] Loss at epoch 36 step 96: [0.01665253], acc: [1.] Loss at epoch 36 step 108: [0.03808357], acc: [1.] Loss at epoch 36 step 120: [0.02877017], acc: [0.984375] Loss at epoch 36 step 132: [0.03208008], acc: [1.] Loss at epoch 36 step 144: [0.02449581], acc: [1.] Loss at epoch 37 step 0: [0.01770885], acc: [1.] Loss at epoch 37 step 12: [0.0464542], acc: [1.] Loss at epoch 37 step 24: [0.0226934], acc: [1.] Loss at epoch 37 step 36: [0.01329492], acc: [1.] Loss at epoch 37 step 48: [0.01974557], acc: [1.] Loss at epoch 37 step 60: [0.01571712], acc: [1.] Loss at epoch 37 step 72: [0.02419647], acc: [1.] Loss at epoch 37 step 84: [0.02948266], acc: [0.984375] Loss at epoch 37 step 96: [0.01232352], acc: [1.] Loss at epoch 37 step 108: [0.02366105], acc: [1.] Loss at epoch 37 step 120: [0.01187477], acc: [1.] Loss at epoch 37 step 132: [0.02855218], acc: [1.] Loss at epoch 37 step 144: [0.05969697], acc: [0.984375] Loss at epoch 38 step 0: [0.01876382], acc: [1.] Loss at epoch 38 step 12: [0.01727666], acc: [1.] Loss at epoch 38 step 24: [0.01147198], acc: [1.] Loss at epoch 38 step 36: [0.00777123], acc: [1.] Loss at epoch 38 step 48: [0.01662749], acc: [1.] Loss at epoch 38 step 60: [0.04357544], acc: [0.984375] Loss at epoch 38 step 72: [0.03622793], acc: [0.984375] Loss at epoch 38 step 84: [0.02586586], acc: [1.] Loss at epoch 38 step 96: [0.02165689], acc: [0.984375] Loss at epoch 38 step 108: [0.0362319], acc: [1.] Loss at epoch 38 step 120: [0.08535068], acc: [0.96875] Loss at epoch 38 step 132: [0.0680019], acc: [0.984375] Loss at epoch 38 step 144: [0.0185735], acc: [1.] Loss at epoch 39 step 0: [0.0207237], acc: [1.] Loss at epoch 39 step 12: [0.02422251], acc: [1.] Loss at epoch 39 step 24: [0.00987013], acc: [1.] Loss at epoch 39 step 36: [0.01326196], acc: [1.] Loss at epoch 39 step 48: [0.00862311], acc: [1.] Loss at epoch 39 step 60: [0.02392986], acc: [0.984375] Loss at epoch 39 step 72: [0.015987], acc: [1.] Loss at epoch 39 step 84: [0.14909726], acc: [0.96875] Loss at epoch 39 step 96: [0.01084465], acc: [1.] Loss at epoch 39 step 108: [0.03847893], acc: [0.984375] Loss at epoch 39 step 120: [0.05880758], acc: [0.96875] Loss at epoch 39 step 132: [0.01939185], acc: [1.] Loss at epoch 39 step 144: [0.01728617], acc: [1.] Loss at epoch 40 step 0: [0.02000599], acc: [1.] Loss at epoch 40 step 12: [0.02437929], acc: [1.] Loss at epoch 40 step 24: [0.01433492], acc: [1.] Loss at epoch 40 step 36: [0.01201964], acc: [1.] Loss at epoch 40 step 48: [0.02156859], acc: [1.] Loss at epoch 40 step 60: [0.01818635], acc: [1.] Loss at epoch 40 step 72: [0.03334533], acc: [0.984375] Loss at epoch 40 step 84: [0.01424423], acc: [1.] Loss at epoch 40 step 96: [0.01449857], acc: [1.] Loss at epoch 40 step 108: [0.0260902], acc: [1.] Loss at epoch 40 step 120: [0.03537105], acc: [1.] Loss at epoch 40 step 132: [0.04467135], acc: [0.984375] Loss at epoch 40 step 144: [0.00684468], acc: [1.] Loss at epoch 41 step 0: [0.00856959], acc: [1.] Loss at epoch 41 step 12: [0.01467489], acc: [1.] Loss at epoch 41 step 24: [0.00849675], acc: [1.] Loss at epoch 41 step 36: [0.023687], acc: [1.] Loss at epoch 41 step 48: [0.01757425], acc: [1.] Loss at epoch 41 step 60: [0.01322003], acc: [1.] Loss at epoch 41 step 72: [0.01735658], acc: [1.] Loss at epoch 41 step 84: [0.05094539], acc: [0.984375] Loss at epoch 41 step 96: [0.03160336], acc: [0.984375] Loss at epoch 41 step 108: [0.03046536], acc: [0.984375] Loss at epoch 41 step 120: [0.01673304], acc: [1.] Loss at epoch 41 step 132: [0.02226919], acc: [1.] Loss at epoch 41 step 144: [0.0173775], acc: [1.] Loss at epoch 42 step 0: [0.00823799], acc: [1.] Loss at epoch 42 step 12: [0.01163502], acc: [1.] Loss at epoch 42 step 24: [0.05233274], acc: [0.984375] Loss at epoch 42 step 36: [0.01162634], acc: [1.] Loss at epoch 42 step 48: [0.01652665], acc: [1.] Loss at epoch 42 step 60: [0.04055682], acc: [1.] Loss at epoch 42 step 72: [0.0074306], acc: [1.] Loss at epoch 42 step 84: [0.01786682], acc: [1.] Loss at epoch 42 step 96: [0.01471556], acc: [1.] Loss at epoch 42 step 108: [0.0290984], acc: [1.] Loss at epoch 42 step 120: [0.00511887], acc: [1.] Loss at epoch 42 step 132: [0.11858401], acc: [0.984375] Loss at epoch 42 step 144: [0.04532209], acc: [0.984375] Loss at epoch 43 step 0: [0.02332157], acc: [0.984375] Loss at epoch 43 step 12: [0.02192533], acc: [1.] Loss at epoch 43 step 24: [0.0114843], acc: [1.] Loss at epoch 43 step 36: [0.02926657], acc: [0.984375] Loss at epoch 43 step 48: [0.0190172], acc: [1.] Loss at epoch 43 step 60: [0.01611589], acc: [1.] Loss at epoch 43 step 72: [0.01435156], acc: [1.] Loss at epoch 43 step 84: [0.01970273], acc: [1.] Loss at epoch 43 step 96: [0.04218911], acc: [0.984375] Loss at epoch 43 step 108: [0.01168987], acc: [1.] ......(AI Studio执行时最后半夜崩溃了，log只保存到这) ```python # 备份模型 !cp checkpoints_models/res3d_model.pdparams checkpoints_models/res3d_model.pdparams.bak ``` ## 2.5 评估 ```python !python eval.py --use_gpu True --weights 'checkpoints_models/res3d_model.pdparams' --config "configs/res3d_L50.txt" ``` [INFO: eval.py: 118]: Namespace(batch_size=1, config='configs/res3d_L50.txt', filelist=None, infer_topk=1, log_interval=1, model_name='res3d', save_dir='./output', use_gpu=True, weights='checkpoints_models/res3d_model.pdparams') {'MODEL': {'name': 'Res3D', 'format': 'pkl', 'num_classes': 101, 'seg_num': 1, 'seglen': 8, 'image_mean': [0.485, 0.456, 0.406], 'image_std': [0.229, 0.224, 0.225], 'num_layers': 50}, 'TRAIN': {'epoch': 50, 'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 36, 'use_gpu': True, 'num_gpus': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/train.list', 'learning_rate': 0.01, 'learning_rate_decay': 0.1, 'l2_weight_decay': 0.0001, 'momentum': 0.9, 'total_videos': 80}, 'VALID': {'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/val.list'}, 'TEST': {'seg_num': 1, 'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/test.list'}, 'INFER': {'short_size': 116, 'target_size': 116, 'num_reader_threads': 1, 'buf_size': 1024, 'batch_size': 1, 'filelist': 'work/data/ucf_101/video/UCF-101_JPG/test.list'}} [INFO: config.py: 68]: ---------------- Valid Arguments ---------------- [INFO: config.py: 70]: MODEL: [INFO: config.py: 72]: name:Res3D [INFO: config.py: 72]: format:pkl [INFO: config.py: 72]: num_classes:101 [INFO: config.py: 72]: seg_num:1 [INFO: config.py: 72]: seglen:8 [INFO: config.py: 72]: image_mean:[0.485, 0.456, 0.406] [INFO: config.py: 72]: image_std:[0.229, 0.224, 0.225] [INFO: config.py: 72]: num_layers:50 [INFO: config.py: 70]: TRAIN: [INFO: config.py: 72]: epoch:50 [INFO: config.py: 72]: short_size:116 [INFO: config.py: 72]: target_size:116 [INFO: config.py: 72]: num_reader_threads:1 [INFO: config.py: 72]: buf_size:1024 [INFO: config.py: 72]: batch_size:36 [INFO: config.py: 72]: use_gpu:True [INFO: config.py: 72]: num_gpus:1 [INFO: config.py: 72]: filelist:work/data/ucf_101/video/UCF-101_JPG/train.list [INFO: config.py: 72]: learning_rate:0.01 [INFO: config.py: 72]: learning_rate_decay:0.1 [INFO: config.py: 72]: l2_weight_decay:0.0001 [INFO: config.py: 72]: momentum:0.9 [INFO: config.py: 72]: total_videos:80 [INFO: config.py: 70]: VALID: [INFO: config.py: 72]: short_size:116 [INFO: config.py: 72]: target_size:116 [INFO: config.py: 72]: num_reader_threads:1 [INFO: config.py: 72]: buf_size:1024 [INFO: config.py: 72]: batch_size:1 [INFO: config.py: 72]: filelist:work/data/ucf_101/video/UCF-101_JPG/val.list [INFO: config.py: 70]: TEST: [INFO: config.py: 72]: seg_num:1 [INFO: config.py: 72]: short_size:116 [INFO: config.py: 72]: target_size:116 [INFO: config.py: 72]: num_reader_threads:1 [INFO: config.py: 72]: buf_size:1024 [INFO: config.py: 72]: batch_size:1 [INFO: config.py: 72]: filelist:work/data/ucf_101/video/UCF-101_JPG/test.list [INFO: config.py: 70]: INFER: [INFO: config.py: 72]: short_size:116 [INFO: config.py: 72]: target_size:116 [INFO: config.py: 72]: num_reader_threads:1 [INFO: config.py: 72]: buf_size:1024 [INFO: config.py: 72]: batch_size:1 [INFO: config.py: 72]: filelist:work/data/ucf_101/video/UCF-101_JPG/test.list [INFO: config.py: 73]: ------------------------------------------------- W0906 07:50:55.139740 301 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 10.1, Runtime API Version: 9.0 W0906 07:50:55.143996 301 device_context.cc:260] device: 0, cuDNN Version: 7.3. 验证集准确率为:0.38435104489326477 # 3 参考说明 此次论文复现参考并引用了PaddlePaddle库和部分3D ResNets论文代码的相关说明，其地址分别为： https://github.com/PaddlePaddle/PaddleHub https://github.com/kenshohara/3D-ResNets-PyTorch