ValueError: Points cannot contain NaN

[kobezy]

您好，请问在运行大些的数据集(如HRSID,RSDD-SAR)时为什么总会出现这个问题：仿佛是模型不太稳定。
Traceback (most recent call last):
File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/train.py", line 199, in run_epoch
loss = criterion(pr_decs, data_dict)
File "/root/miniconda3/envs/gagale/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/loss.py", line 126, in forward
wh_loss, iou_loss = self.L_wh(pr_decs['wh'], gt_batch['reg_mask'], gt_batch['ind'], gt_batch['wh'])#wh_loss 和 iou_loss 分别代表宽度（width）和高度（height）的损失，由 IoUWeightedSmoothL1Loss 计算得出。
File "/root/miniconda3/envs/gagale/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/polar.py", line 453, in forward
ious_all_lists = self._calculate_ious(output, mask, ind, target)
File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/polar.py", line 429, in _calculate_ious
pred_bboxes = self._polar_to_bboxes(valid_pred) # [num_obj, 8]
File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/polar.py", line 354, in _polar_to_bboxes
mbb = MinimumBoundingBox(target_pts)
File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/MBB.py", line 109, in MinimumBoundingBox
hull_ordered = [points[index] for index in ConvexHull(points).vertices]
File "qhull.pyx", line 2431, in scipy.spatial.qhull.ConvexHull.init
File "qhull.pyx", line 283, in scipy.spatial.qhull._Qhull.init
ValueError: Points cannot contain NaN

[flyingshan]

File "/home/Learning-Polar-Encodings-For-Arbitrary-Oriented-Ship-Detection-In-SAR-Images-master/polar.py", line 429, in _calculate_ious
pred_bboxes = self._polar_to_bboxes(valid_pred) # [num_obj, 8]

看起来是极坐标点转直角坐标点之后出现了null值的点，您可以debug看一下这里valid_pred里面是不是有一些有问题的输出

[Untilseeuu]

我也遇到了类似的问题