lornatang/csnln-pytorch

PyTorch implemnts `Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining` paper.

Hi, my dear friend, I am committed to promoting the development of GAN.

csnln-pytorch's Issues

PSNR-inf and SSIM nan

Excuse me, PSNR-inf and SSIM nan appear when I run the first 10 epoch. What should I do?

Pretrained model is not working

Hello,

When I tried to use your pre trained model, I have face this error.

validation mode

results/CSNLN_x3/best.pth.tar
Build CSNLN model successfully.
Traceback (most recent call last):
  File "validate.py", line 95, in <module>
    main()
  File "validate.py", line 34, in main
    model.load_state_dict(checkpoint["state_dict"])
KeyError: 'state_dict'

config.py

import random
from traceback import print_tb

import numpy as np
import torch
from torch.backends import cudnn

# Random seed to maintain reproducible results
random.seed(0)
torch.manual_seed(0)
np.random.seed(0)
# Use GPU for training by default
device = torch.device("cuda", 0)
# Turning on when the image size does not change during training can speed up training
cudnn.benchmark = True
# Image magnification factor
upscale_factor = 3
# Current configuration parameter method
mode = "valid"
# Experiment name, easy to save weights and log files
exp_name = "CSNLN_x" + str(upscale_factor)

if mode == "train":
    # Dataset
    train_image_dir = f"data/DIV2K/CSNLN/train"
    valid_image_dir = f"data/DIV2K/CSNLN/valid"
    test_lr_image_dir = f"data/Set5/LRbicx{upscale_factor}"
    test_hr_image_dir = f"data/Set5/GTmod12"

    image_size = int(upscale_factor * 48)
    batch_size = 16
    num_workers = 4

    # Incremental training and migration training
    start_epoch = 0
    resume = ""

    # Total num epochs
    epochs = 50

    # Adam optimizer parameter
    model_lr = 1e-4
    model_betas = (0.9, 0.999)

    # StepLR scheduler parameter
    lr_scheduler_step_size = 15
    lr_scheduler_gamma = 0.5

    print_frequency = 100

if mode == "valid":
    print("validation mode")
    # Test data address
    lr_dir = f"data/Set5/LRbicx{upscale_factor}"
    sr_dir = f"results/test/{exp_name}"
    hr_dir = f"data/Set5/GTmod12"
    
    model_path = f"results/{exp_name}/best.pth.tar"
    print(model_path)

how to train x4model

hi! the sever show out of memory when i train x4model. i should how to set bath_size and patch_size? my gpu gtx 24g