search

wei-tim / YOWO

You Only Watch Once: A Unified CNN Architecture for Real-Time Spatiotemporal Action Localization
Other
846 stars 158 forks source link

How to detect certain actions in my own video data #17

Closed west-i closed 4 years ago

west-i commented 4 years ago

What steps do I need to take if I want to detect a certain action in my video data.

wei-tim commented 4 years ago

@west-i Hi, thanks for your interest. First you have to check whether the action that you are going to detect is in the datasets (JHMDB-21 or UCF101-24). If so, you can directly use our pretrained model to detect this specific action. Otherwise you can use our method to train your own model on your specific dataset.

west-i commented 4 years ago

@wei-tim Thanks you for your answer.

The action that i want to detect is not in the datasets(JHMDB-21 or UCF101-24). I have the following questions：

1.How should I annotate the new action in the video to train my detection model. 2.What needs to be modified next to match my data?

Beginners in this area, hope to get your answers, thank you very much!

abhigoku10 commented 4 years ago

@okankop @west-i if the actiion is present in datasets(JHMDB-21 or UCF101-24). which set of file / the process should be follow to run on the video data

okankop commented 4 years ago

@west-i

  1. you should annotate your data similar to UCF101-24 annotations.
  2. You just need to change the number of classes in the network architecture and you are ready to train.
okankop commented 4 years ago

@abhigoku10 you can simply follow the test procedure applied in the training. Please investigate the "def test(epoch)" function in file https://github.com/wei-tim/YOWO/blob/master/train.py

west-i commented 4 years ago

@okankop Thanks you for your answer.

Is there a labeling tool I can use to label my videos?

okankop commented 4 years ago

@west-i Labelme (https://github.com/wkentaro/labelme) is a very nice labelling tool. You can use that.

okankop commented 4 years ago

I believe issue is resolved!