语言:
🇨🇳
🇺🇸
«TSN»复现了论文Temporal Segment Networks提出的视频分类模型
Temporal Segment Networks: Towards Good Practices for Deep Action Recognition是视频分类任务中的经典实现
通过requirements.txt
安装运行所需依赖
$ pip install -r requirements.txt
处理数据时需要额外安装denseflow,可以在innerlee/setup中找到安装脚本
首先设置GPU
和当前位置
$ export CUDA_VISIBLE_DEVICES=1
$ export PYTHONPATH=.
- 训练
# 训练UCF101
# 单GPU
$ python tools/train.py --config_file=configs/tsn_r50_ucf101_rgb_224x3_seg.yaml
# 多GPU
$ python tools/train.py \
--config_file=configs/tsn_r50_ucf101_rgb_224x3_seg.yaml \
--eval_step=1000 \
--save_step=1000 \
-g=<N>
- 测试
# 单模态测试
$ python tools/test.py <config_file> <pth_file>
$ python tools/test.py configs/tsn_r50_ucf101_rgb_224x3_seg.yaml outputs/tsn_r50_ucf101_rgb_224x3_seg.pth
# 多模态融合测试 - RGB + RGBDiff
$ python tools/fusion.py <rgb_config_file> <rgb_pth_file> <rgbdiff_config_file> <rgbdiff_pth_file>
$ python tools/fusion.py \
configs/tsn_r50_ucf101_rgb_224x3_seg.yaml \
outputs/tsn_r50_ucf101_rgb_224x3_seg.pth \
configs/tsn_r50_ucf101_rgbdiff_224x3_seg.yaml \
outputs/tsn_r50_ucf101_rgbdiff_224x3_seg.pth
- zhujian - Initial work - zjykzj
@misc{wang2016temporal,
title={Temporal Segment Networks: Towards Good Practices for Deep Action Recognition},
author={Limin Wang and Yuanjun Xiong and Zhe Wang and Yu Qiao and Dahua Lin and Xiaoou Tang and Luc Van Gool},
year={2016},
eprint={1608.00859},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
欢迎任何人的参与!打开issue或提交合并请求。
注意:
GIT
提交,请遵守Conventional Commits规范- 语义版本化,请遵守Semantic Versioning 2.0.0规范
README
编写,请遵守standard-readme规范
Apache License 2.0 © 2020 zjykzj