So I was going through the gt_creatorfunction inside the create_gtscript.
I checked the output of gt_tensorusing a few targets.
I first tried it using a target that had only 1 bounding box. The shape of gt_tensor is torch.Size([128, 384, 17]).
Then I tried using a target that had 10 bounding boxes. But the shape of gt_tensor remained torch.Size([128, 384, 17])
This is slightly confusing. As I thought there should be N gt_tensors for N bounding boxes. How does the shape of the gt_tensor remain the same irrespective of the number of bounding boxes?
Edit: I understood this part.
However, (down below)
NOTE: The model gives top 100 predictions irrespective of the number of bounding boxes. So for example, if the target contains 1 bounding box the model gives 100 predictions, and even if the target has 10 bounding boxes, it still gives 100 predictions? How does this work?
Hi @yjh0410 ,
So I was going through the
gt_creatorfunction inside thecreate_gtscript. I checked the output ofgt_tensorusing a few targets.torch.Size([128, 384, 17]).torch.Size([128, 384, 17])This is slightly confusing. As I thought there should be N gt_tensors for N bounding boxes. How does the shape of the gt_tensor remain the same irrespective of the number of bounding boxes?
Edit: I understood this part.
However, (down below)
NOTE: The model gives top 100 predictions irrespective of the number of bounding boxes. So for example, if the target contains 1 bounding box the model gives 100 predictions, and even if the target has 10 bounding boxes, it still gives 100 predictions? How does this work?