Question about converting custom dataset to coco api

[kangkang59812]

vision/references/detection/coco_utils.py

Line 163 in a44d55d

bboxes[:, 2:] -= bboxes[:, :2]

If the box is [3,10,6,20](xyxy format),the converted box should be [3,10,4,11]. I think this code should be added 1. Because there are 4 pixels between [3,6] and 11 pixels between [10,20]. It actually computes the pixels in grid.
May be the original computation of the area need to do this as well. Such as this tutorial, https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
area = (boxes[:, 3] - boxes[:, 1]) * (boxes[:, 2] - boxes[:, 0])

[fmassa]

Hi,

This has been discussed in #826 (comment) , and the addition of +1 was removed from torchvision, and we consider box coordinates in terms of continuous coordinates, in which case the +1 is not necessary.

Let me know if you have further questions