mmaction2训练微调模型

使用预训练模型
在configs/base/default_runtime.py中
将load_from的默认值改为

load_from = 'checkpoints/tsn_r50_1x1x3_100e_kinetics400_rgb/tsn_r50_1x1x3_100e_kinetics400_rgb_20200614-e508be42.pth'  # model path can be found in model zoo

加载配置文件
 "args": [         "configs/recognition/tsn/tsn_r50_1x1x3_100e_kinetics400_rgb.py" ]

数据集准备惨参考链接
https://github.com/open-mmlab/mmaction2/blob/master/docs/en/data_preparation.md

https://github.com/open-mmlab/mmaction2/blob/master/configs/recognition/tsn/README.md

下载完数据进行裁剪

python ../resize_videos.py ../../../data/${DATASET}/videos_train/ ../../../data/${DATASET}/videos_train_256p_dense_cache --dense --level 2