northeastsquare
commented
5 years ago While with latest pytorch, use https://github.com/t-vi/warp-ctc/, there's some error, t-vi said, he will fix it some. While with latest pytorch, but with pytorch's own ctc-loss, use batch size 64, back prop will have many nan.
jiangxiluning
Hello, I run the code with some error as follows, so may I know your version of pytorch? `Memory Usage: CUDA: 0 Allocated: 117.5244140625 MB Cached: 125.875 MB
Traceback (most recent call last): File "train.py", line 76, in
main(config, args.resume)
File "train.py", line 50, in main
trainer.train()
File "/home/sam/TextDetection/FOTS.PyTorch/FOTS/base/base_trainer.py", line 78, in train
result = self._train_epoch(epoch)
File "/home/sam/TextDetection/FOTS.PyTorch/FOTS/trainer/trainer.py", line 59, in _train_epoch
for batch_idx, gt in enumerate(self.data_loader):
File "/usr/local/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 314, in next
batch = self.collate_fn([self.dataset[i] for i in indices])
File "/home/sam/TextDetection/FOTS.PyTorch/FOTS/data_loader/datautils.py", line 696, in collate_fn
a = torch.from_numpy(img[i])
RuntimeError: PyTorch was compiled without NumPy support`