fasterrcnn-pytorch-training-pipeline

fasterrcnn-pytorch-training-pipeline copied to clipboard

I often use this repo as a point of reference.

boxes = [] 
boxes = torch.as_tensor(boxes, dtype=torch.float32)

I don't believe the above would create the required shape to be provided to Faster RCNN in the case of an empty frame. See https://stackoverflow.com/questions/66063046/how-to-train-faster-rcnn-on-dataset-including-negative-data-in-pytorch .

Does your dataset / model cope with empty frames in some other way? Just trying to find what's considered sensible practice.

https://github.com/pytorch/vision/issues/1598

@GeorgePearse Yes, it handles empty frames. This part in datasets.py handles it.