OpenGVLab / UniFormerV2

[ICCV2023] UniFormerV2: Spatiotemporal Learning by Arming Image ViTs with Video UniFormer
https://arxiv.org/abs/2211.09552
Apache License 2.0
294 stars 19 forks source link

feature extraction #7

Closed M-Mokari closed 1 year ago

M-Mokari commented 1 year ago

I want to use your method as extracted feature such as I3D for my special application. could you help me? Do you have extracted feature specially on activity net dataset?

Andy1621 commented 1 year ago

Sorry. I do not have extracted feature. I am conducting model transfer in Hugging Face and MMAction2, which will provide a feature extractor.

BTW, if you want to extract the feature by yourself, you can use the model code here. Then, you can simply use the pretrained model I have provided, and merge the model in your progress for feature extraction, like MMAction2. You may need to set the sampling rate, and input resolution by yourself.

M-Mokari commented 1 year ago

Thanks a lot. It could be useful.

yhl2018 commented 1 year ago

I want to use your method as extracted feature too! could you help me? Modify the model to look like swin transformer or like MMAction2 backbone