Kai Hu

from torchvision

You can refer to https://github.com/open-mmlab/mmaction2/blob/master/configs/localization/bmn/README.md. The ` init_recognizer` and `inference_recognizer` functions are used for classification tasks.

Did not try this method. But the deploy of spatial-temporal detection models will be supported soon. Please stay tuned.

@ysy070100 You can refer to ``Train Custom Classes From Ava Dataset`` the section at https://github.com/open-mmlab/mmaction2/blob/master/configs/detection/ava/README.md.

You can download the pre-train checkpoint trained on AVA from our repo, and fine-tune the model with your data. You can also freeze the network backbone and train the classifier...

@ysy070100 You can look at the `Modify Training Schedule` section here `tutorial 2: finetuning models`: https://mmaction2.readthedocs.io/en/latest/tutorials/2_finetune.html#modify-training-schedule

If you need any further discussions, feel free to reopen the issue. Thanks!

I did run action detection models on AVA-Kinetics. The changes on codes are not very complicated, but the data processing is rather painful. Many videos in the Kinetics 700 dataset...

Hi @kennymckormick I can implement the AVA-Kinetics dataset and training. Training on AVA-Kinetics indeed helps the performance on AVA a lot. For the AVA-Kinetics, I plan to use YouTube-dl to...

@irvingzhang0512 My first attempt is the CVDF repo. One issue is that they have K400, K600 and K700-2020 but not K700. Some videos are different for K700-2020 and K700 which...