fix a small bug

[Yzichen]

If k>0 is used as the condition, cur_gt.shape[0] will always be greater than 0. For some scenes without targets, cur_gt will be a zero tensor with shape (1, 8), which affects recall_dict['gt'], As a result, it affects the calculation of recall.

[sshaoshuai]

Thanks for the PR!

Yes, you are right. The first part is a small error.

In the second part of your PR, I require cur_gt.shape[0] > 0 to avoid some unnecessary errors since some codes in the repo may require each scene has at least one GT, or it will cause some errors.
For example, the line 96 will still fake a zero GT in line 96 to avoid the error in the target assigning process:

OpenPCDet/pcdet/models/roi_heads/target_assigner/proposal_target_layer.py

Lines 92 to 96 in 66c7ddb

	k = cur_gt.__len__() - 1
	while k >= 0 and cur_gt[k].sum() == 0:
	k -= 1
	cur_gt = cur_gt[:k + 1]
	cur_gt = cur_gt.new_zeros((1, cur_gt.shape[1])) if len(cur_gt) == 0 else cur_gt

Nevertheless, thanks for your PR and I will accept it.