Training the action model from scratch

[albertaparicio]

I was looking into this project and I was wondering if there is a supported way for training the Action Recognition model from scratch. I have not seen any mention of it in the documentation nor I have seen any script that indicates this is supported.

Any pointers in this matter are appreciated. Thank you

[ycxioooong]

Hi there,

We do not support training and will not support it in the future.

But you could use mmaction for training action models.
Since our framework follows the same API design as mmaction,
you could very easily transfer the pre-trained models from mmaction to movienet.tools and extract action features using movienet.tools.

Besides, currently mmaction already supports a wide range of methods on all the action understanding tasks, including action recognition, action temporal detection, and spatial-temporal action detection.

[ycxioooong]

By the way, if you want to train person detection models by yourself, please refer to mmdetection.