Use RWF-2000 to train TSM

mit-han-lab / temporal-shift-module

[ICCV 2019] TSM: Temporal Shift Module for Efficient Video Understanding

https://arxiv.org/abs/1811.08383

MIT License

2.07k stars 417 forks source link

Use RWF-2000 to train TSM #154

Open bujianyiwang opened 4 years ago

bujianyiwang commented 4 years ago

How can I deal with RWF-2000 when following the instruction of Data Preparation? My dataset is RWF-2000 which contatins train and val folder,label is only Fight and NonFight.

hoangminhtoan commented 3 years ago

Hi @bujianyiwang. You can modify these files for rwf-2000. I created 3 folders:

RWF-2000 (contains original videos. e.g. xyz.avi), I remove all special characters (Chinese, Russian, space, etc.) to avoid error before using ffmeg command
RWF-2000-raw-frames (contains original video names after converting video to its corresponding frames. e.g xyz.avi -> xyz/00001.jpg, xyz/00002.jpg, etc.)
RWF-2000-frames (convert folder frames into integer base as the same format in something2: e.g: xyz -> 1, abc -> 2, etc.) rwf_datagen.zip

lxzatwowone1 commented 3 years ago

RWF-2000效果不错的

bujianyiwang commented 2 years ago

@hoangminhtoan Do you have git repo to follow? Most confused thing is the dataset preprocess and dataloader, especially on my own custom dataset. I just use the open dataset(RWF-2000) closed to my applied scene to put through a route for custom dataset. THX!

bujianyiwang commented 2 years ago

@lxzatwowone1 有repo可以follow吗？