COCO keypoints - Githubissues

NVIDIA / DALI

A GPU-accelerated library containing highly optimized building blocks and an execution engine for data processing to accelerate deep learning training and inference applications.

https://docs.nvidia.com/deeplearning/dali/user-guide/docs/index.html

Apache License 2.0

5.01k stars 610 forks source link

COCO keypoints #2868

Open viotemp1 opened 3 years ago

viotemp1 commented 3 years ago

Hello,

I want to use DALI to retrain some models for human pose detection - with COCO 2017 I see nvidia.dali.fn.readers.coco support images, bounding_boxes, labels, ((polygons, vertices) | (pixelwise_masks)), (image_ids). Is there any way to add keypoints to this reader? Regards,

JanuszL commented 3 years ago

Hi @viotemp1,

It should be possible. We don't have that on our roadmap yet. If you are willing to contribute to DALI we will be more than happy to review and accept any PR with such functionality. What you should start from are:

coco_loader.cc and extend ParseAnnotations to parse keypoints as well and ParseJsonAnnotations to store it in an appropriate strucutre
coco_reader_op.cc and add an option to return keypoints as an optional output

viotemp1 commented 3 years ago

Hello,

My knowledge in c++ is pretty limited (better to say - not used c++ for long time), but sure I can try. I'll fork the repo and start to look into it. I might need help from time to time ... Regards,