关于loss的疑问

对于SparseOcc/models/loss_utils.py中291行的 loss_classes 计算我有一些疑惑

tgt_class = class_gt[b]
tgt_mask = (tgt_mask.unsqueeze(-1) == torch.arange(num_instances).to(mask_gt.device))
tgt_mask = tgt_mask.permute(1, 0)
            
src_idx, tgt_idx = indices[b]
src_mask = mask_pred[b][src_idx]   # [M, N], M is number of gt instances, N is number of remaining voxels
tgt_mask = tgt_mask[tgt_idx]   # [M, N]
src_class = class_pred[b]   # [Q, CLS]
            
# pad non-aligned queries' tgt classes with 'no class'
pad_tgt_class = torch.full(
       (src_class.shape[0], ), self.num_classes - 1, dtype=torch.int64, device=class_pred.device
)   # [Q]
pad_tgt_class[src_idx] = tgt_class

为什么这里的 tgt_class 不用加 tgt_class[tgt_idx] 呢，而是直接 tgt_class = class_gt[b]

The issue points to SparseOcc/models/loss_utils.py around line 291. Start by tracing class_gt, indices, and tgt_idx into the loss_classes calculation and checking their shapes and ordering. Done means the intended indexing behavior is explained in the issue or the implementation is corrected if the ordering is inconsistent.