frame extraction about two-stream-pytorch

[bryanyzhu]

Hi you can either use opencv or ffmpeg to extract the frames. I suggest using opencv.

The detailed instructions are here. You can follow the sample in "saving a video" section. Hope this helps.

[chandnikathuria1992]

actually i already have extracted frames using FFmpeg
with folder structure as UCF101\jpg\classname\videoname\image_0001.jpg
now i want to test ur spatial stream.So first i need to run build_file_list.py. For this what should be folder structure for video frames?

[bryanyzhu]

Hi, if you already extract your frames, then maybe you can manually write your own script to generate the train/test file list.

The train file list should look like this. And when your train your model, you use "UCF101\jpg\classname" as DATA_PATH (as you said, this is your folder structure).

Note that, our image name pattern is default to be img_00001.jpg as written here. You can pass your own pattern to the data loading function, or simply change line 133 to image_%04d.jpg to fit your case.

If you want to do test, then the test file list should look like this.

I am sorry the code is not well organized, don't have time to unify it recently.

[chandnikathuria1992]

Thanks !

…

On 17 Feb 2018 10:40 pm, "Yi Zhu" ***@***.***> wrote: Hi, if you already extract your frames, then maybe you can manually write your own script to generate the train/test file list. The train file list should look like this <https://github.com/bryanyzhu/two-stream-pytorch/blob/master/datasets/settings/ucf101/train_rgb_split1.txt>. And when your train your model, you use "UCF101\jpg\classname" as DATA_PATH (as you said, this is your folder structure). Note that, our image name pattern is default to be img_00001.jpg as written here <https://github.com/bryanyzhu/two-stream-pytorch/blob/master/datasets/ucf101.py#L129-L135>. You can pass your own pattern to the data loading function, or simply change line 133 to image_%04d.jpg to fit your case. If you want to do test, then the test file list should look like this <https://github.com/bryanyzhu/two-stream-pytorch/blob/master/scripts/eval_ucf101_pytorch/spatial_testlist01_with_labels.txt> . I am sorry the code is not well organized, don't have time to unify it recently. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#16 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Adg6KR-VZ2b8JUWD9lLdUDhp46DRyhMxks5tVwf5gaJpZM4SJNSm> .