search

liqilei / GMFN

PyTorch code for our paper "Gated Multiple Feedback Network for Image Super-Resolution" (BMVC2019)
70 stars 13 forks source link

train error size not match #2

Closed yja1 closed 5 years ago

yja1 commented 5 years ago

CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_GMFN.json I use celeba dataset train

===> Training Epoch: [1/1000]... Learning Rate: 0.000200 Epoch: [1/1000]: 0%| | 0/251718 [00:00<?, ?it/s] Traceback (most recent call last): File "train.py", line 131, in main() File "train.py", line 69, in main iter_loss = solver.train_step() File "/exp_sr/SRFBN/solvers/SRSolver.py", line 104, in train_step loss_steps = [self.criterion_pix(sr, split_HR) for sr in outputs] File "/exp_sr/SRFBN/solvers/SRSolver.py", line 104, in loss_steps = [self.criterion_pix(sr, split_HR) for sr in outputs] File "/toolscnn/env_pyt0.4_py3.5_awsrn/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in call result = self.forward(*input, **kwargs) File "/toolscnn/env_pyt0.4_py3.5_awsrn/lib/python3.5/site-packages/torch/nn/modules/loss.py", line 87, in forward return F.l1_loss(input, target, reduction=self.reduction) File "/toolscnn/env_pyt0.4_py3.5_awsrn/lib/python3.5/site-packages/torch/nn/functional.py", line 1702, in l1_loss input, target, reduction) File "/toolscnn/env_pyt0.4_py3.5_awsrn/lib/python3.5/site-packages/torch/nn/functional.py", line 1674, in _pointwise_loss return lambd_optimized(input, target, reduction) RuntimeError: input and target shapes do not match: input [16 x 3 x 192 x 192], target [16 x 3 x 48 x 48] at /pytorch/aten/src/THCUNN/generic/AbsCriterion.cu:12

liqilei commented 5 years ago

You are suggested to check the path of your HR image at HERE.

yja1 commented 5 years ago

there is some limitation of train LR HR width and height ? at least big than ? pixel

liqilei commented 5 years ago

For training, the image should be larger than the training patch size, i.e. 48x48.

Your error input and target shapes do not match: input [16 x 3 x 192 x 192], target [16 x 3 x 48 x 48] indicates that your feed the incorrect HR image. In other words, your LR image is 48x48, with up scale factor of 4, your HR target should be 192x192, while the given one is 48x48. Your are suggested to check the path of your HR image, and the shape of you HR images.