Ava Dataset Format Preparation for Custom Dataset

[Naeem Ullah Khan]

Thanks for providing such an amazing and comprehensive dataset for activity recognition. But my question is related to Ava dataset format preparation for a custom dataset. Can you guide me on what are pre-processing step that needs to do get format like AVA Dataset? What are the available tools which help me to achieve that format? Please guide me about this because I have not found any help regarding custom dataset formation for AVA. Thanks 

My dataset has three classes and it has a structure like this 

Dataset/
├── Class-A
│   └── insp.mp4
├── Class-B
│   └── kep.mp4
└── Class-C
    └── tak.mp4

[David Ross]

Hi Naeem,

Thanks for your interest in AVA. We train our AVA models in TensorFlow, so we convert the video data to tensorflow.SequenceExample protocol buffers, using the MediaSequence module of MediaPipe. To identify person bounding boxes in the video, we use a person detector model from the TensorFlow Object Detection API, which has pre-trained models in their Model Zoo.

Do other people have end-to-end tools for working with AVA that they would recommend?

Cheers,

David Ross

Google Research

--
You received this message because you are subscribed to the Google Groups "AVA Dataset Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ava-dataset-us...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ava-dataset-users/1b72e584-05ea-416b-aeeb-77d4d3155eb0%40googlegroups.com.

[Naeem Ullah Khan]

I do not find any process or method which helps to make format like AVA Format. I find any recommended tool for the annotation of action videos. Please guide me on what actually I need to do the steps which help to achieve my custom dataset into AVA-Dataset format. Briefly guide me, I will be very thankful