hsi_denoising_test.py

import torch
import torch.nn as nn
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from torchvision.transforms import Compose
import os
import argparse
from utility import *
from hsi_setup import Engine, train_options, make_dataset
import time

if __name__ == '__main__':
    """Training settings"""
    parser = argparse.ArgumentParser(
        description='Hyperspectral Image Denoising (Complex noise)')
    opt = train_options(parser)
    print(opt)
    

    """Setup Engine"""
    engine = Engine(opt)

    """Dataset Setting"""
    
    HSI2Tensor = partial(HSI2Tensor, use_2dconv=engine.net.use_2dconv)

   
    target_transform = HSI2Tensor()



    """Test-Dev"""
    basefolder = opt.testdir
    
    
    mat_datasets = [MatDataFromFolder(
        basefolder) ]

    if not engine.get_net().use_2dconv:
        mat_transform = Compose([
            LoadMatHSI(input_key='input', gt_key='gt',
                    transform=lambda x:x[ ...][None], needsigma=False),
        ])
    else:
        mat_transform = Compose([
            LoadMatHSI(input_key='input', gt_key='gt', needsigma=False),
        ])

    mat_datasets = [TransformDataset(mat_dataset, mat_transform)
                    for mat_dataset in mat_datasets]

 
    mat_loaders = [DataLoader(
        mat_dataset,
        batch_size=1, shuffle=False,
        num_workers=1, pin_memory=opt.no_cuda
    ) for mat_dataset in mat_datasets]        

    base_lr = opt.lr
    epoch_per_save = 5
    adjust_learning_rate(engine.optimizer, opt.lr)

    # from epoch 50 to 100
    engine.epoch  = 0
    
    strart_time = time.time()
    engine.test(mat_loaders[0], basefolder)
    end_time = time.time()
    test_time = end_time-strart_time
    print('cost-time: ',(test_time/50))