Error when training coco @batch 1473, epoch 0

[CaptainEven]

Epoch 0/1000, Batch 1473/7329 | Losses: x 0.126482, y 0.129091, w 0.692879, h 0.779148, conf 0.268520, cls 1.789521, total 3.785640
Traceback (most recent call last):
File "/mnt/diskb/even/yolov3_pytorch/train.py", line 97, in
for batch_i, (_, imgs, targets) in enumerate(dataloader):
File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 264, in next
batch = self.collate_fn([self.dataset[i] for i in indices])
File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 264, in
batch = self.collate_fn([self.dataset[i] for i in indices])
File "/mnt/diskb/even/yolov3_pytorch/utils/datasets.py", line 73, in getitem
h, w, _ = img.shape
ValueError: not enough values to unpack (expected 3, got 0)

[alexandru-dinu]

Because img is a numpy array that wraps the returned PIL Image, I am guessing that None is returned, and so np.array(None).shape is (), hence the got 0.

[CaptainEven]

@alexandru-dinu the training is fine from start, and the error happened @Batch 1473....i don't understand

[chijan-nh]

I got similar issue in detect.py.
The attached data/samples can be run, however, if grayscale images are given as data/samples, then "not enough values to unpack" error is reproduced.
The code h,w,_=img.shape caused this error because the grayscale image does not have channels, i.e. , len(img.shape)==2.

I think the @Batch 1473 's images are broken (including None). The #36 issue may be caused by the dataset. Until @Batch 1472, the dataset is normal, I think.
My dirty hack is (1) get len(img.shape) and (2) use if divided; if len(img.shape)!=3, then this batch is ignored.

[eriklindernoren]

This was caused by a grayscale image in the dataset. It's fixed in master.