SomeThing Weird About The Yolact SegHead

[zzzbosss1997]

Hi, When I look the code in mmdet/models/dense_heads/yolact_head.py/

@HEADS.register_module() class YOLACTSegmHead(nn.Module): def get_targets(self, segm_pred, gt_masks, gt_labels): """Compute semantic segmentation targets for each image.

    Args:
        segm_pred (Tensor): Predicted semantic segmentation map
            with shape (num_classes, H, W).
        gt_masks (Tensor): Ground truth masks for each image with
            the same shape of the input image.
        gt_labels (Tensor): Class indices corresponding to each box.

    Returns:
        Tensor: Semantic segmentation targets with shape
            (num_classes, H, W).
    """
    if gt_masks.size(0) == 0:
        return None
    num_classes, mask_h, mask_w = segm_pred.size()
    with torch.no_grad():
        downsampled_masks = F.interpolate(
            gt_masks.unsqueeze(0), (mask_h, mask_w),
            mode='bilinear',
            align_corners=False).squeeze(0)
        downsampled_masks = downsampled_masks.gt(0.5).float()
        segm_targets = torch.zeros_like(segm_pred, requires_grad=False)
        for obj_idx in range(downsampled_masks.size(0)):
            segm_targets[gt_labels[obj_idx] - 1] = torch.max(
                segm_targets[gt_labels[obj_idx] - 1],
                downsampled_masks[obj_idx])
        return segm_targets

Why using gt_labels[obj_idx]-1? When using coco datasets provided by mmdet, the category label has converted to [0,79], I think there is no need to -1 here. Can someone explain it?

[hhaAndroid]

@zzzbosss1997 Hi, Do you have any questions now?