GitHub - KenyonZhao233/TSN_pytorch: In 2018 Deecamp, Deight(Group 8), our team have reproduced and revised this TSN model based on Pytorch

环境配置参考https://github.com/milkcat0904/temporal-segment-network-pytorch
替换.py文件
train: python3.5 main.py ucf101 Flow ../../action_dataset/action5.0/train_opencv_flow/labels_txt ../../action_dataset/action5.0/test_opencv_flow/labels_txt --arch BNInception --num_segments 3 --gd 20 --lr 0.001 --lr_steps 30 60 --epochs 80 -b 128 -j 8 --dropout 0.8 --snapshot_pref action5.0__opencv__flow_model_best.pth.tar --gpus 0 1
test： python3.5 test_models.py ucf101 RGB ../../action_dataset/action5.0/test_opencv_flow/labels_txt action5.0__opencv_rgb_model_best.pth.tar --arch BNInception --save_scores action5.0_test_rgb.npz --gpu 1 -j 1

5.请使用pytorch0.3，pytorch0.4会发生一系列包括加载模型维度不同等问题

6.Flow推荐使用opencv原生取代原作者的Denseflow

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
dataset.py		dataset.py
eval_scores.py		eval_scores.py
eval_scores_py3.py		eval_scores_py3.py
eval_zky_scores.py		eval_zky_scores.py
main.py		main.py
models.py		models.py
opts.py		opts.py
readme.md		readme.md
test_models.py		test_models.py
test_slipt.py		test_slipt.py
test_zmd_models.py		test_zmd_models.py
transforms.py		transforms.py
zmd_dataset.py		zmd_dataset.py

Provide feedback