search

Tramac / Fast-SCNN-pytorch

A PyTorch Implementation of Fast-SCNN: Fast Semantic Segmentation Network
Apache License 2.0
379 stars 93 forks source link

Support of testing for any size image #34

Open rose-jinyang opened 4 years ago

rose-jinyang commented 4 years ago

Hello How are you? Thanks for contributing this project. It seems that the script demo.py can not process any size of image. Could u revise the script so that it can process any size of image? Thanks.

Tramac commented 4 years ago

There is no limit to the image size in the demo.py.

rose-jinyang commented 4 years ago

Hi I used the size 875x583 of image. Please check this issue.

Traceback (most recent call last): File "E:/Jin/Internet_Task/Jobs/2020/2020.08.31-U2_Net/Fast-SCNN-pytorch-master/demo.py", line 57, in demo() File "E:/Jin/Internet_Task/Jobs/2020/2020.08.31-U2_Net/Fast-SCNN-pytorch-master/demo.py", line 49, in demo outputs = model(image) File "C:\Users\admin\Anaconda3\envs\py37_tf2.x\lib\site-packages\torch\nn\modules\module.py", line 532, in call result = self.forward(*input, *kwargs) File "E:\Jin\Internet_Task\Jobs\2020\2020.08.31-U2_Net\Fast-SCNN-pytorch-master\models\fast_scnn.py", line 37, in forward x = self.feature_fusion(higher_res_features, x) File "C:\Users\admin\Anaconda3\envs\py37_tf2.x\lib\site-packages\torch\nn\modules\module.py", line 532, in call result = self.forward(input, **kwargs) File "E:\Jin\Internet_Task\Jobs\2020\2020.08.31-U2_Net\Fast-SCNN-pytorch-master\models\fast_scnn.py", line 213, in forward out = higher_res_feature + lower_res_feature RuntimeError: The size of tensor a (110) must match the size of tensor b (112) at non-singleton dimension 3

Thanks

rose-jinyang commented 4 years ago

Hi I did padding so that the width and height of image can be multiples of 32. So the issue disappeared but the result is not normal. This is the padded input image. 1

That is the output image. Esha-gupta

How can I understand this? Thanks

milad-4274 commented 3 years ago

same problem. did you find any solution? thanks

Linda-L commented 3 years ago

There was no such problem during the first training. The same data set for the second training is reported as follows：

Traceback (most recent call last): File "/home/lyl/000_Code/Fast-SCNN/train.py", line 200, in <module> trainer.train() File "/home/lyl/000_Code/Fast-SCNN/train.py", line 127, in train for i, (images, targets) in enumerate(self.train_loader): File "/home/lyl/anaconda3/envs/fastscnn1.0/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 615, in __next__ batch = self.collate_fn([self.dataset[i] for i in indices]) File "/home/lyl/anaconda3/envs/fastscnn1.0/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 615, in <listcomp> batch = self.collate_fn([self.dataset[i] for i in indices]) File "/home/lyl/000_Code/Fast-SCNN/data_loader/cityscapes.py", line 93, in __getitem__ img, mask = self._sync_transform(img, mask) File "/home/lyl/000_Code/Fast-SCNN/data_loader/cityscapes.py", line 141, in _sync_transform img = img.resize((ow, oh), Image.BILINEAR) File "/home/lyl/anaconda3/envs/fastscnn1.0/lib/python3.6/site-packages/PIL/Image.py", line 1923, in resize return self._new(self.im.resize(size, resample, box)) ValueError: height and width must be > 0