## Restormer: Efficient Transformer for High-Resolution Image Restoration
## Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang
## https://arxiv.org/abs/2111.09881


import torch
import torch.nn.functional as F
import os
from runpy import run_path
from skimage import img_as_ubyte
import cv2
from tqdm import tqdm
import argparse

parser = argparse.ArgumentParser(description='Test Restormer on your own images')
parser.add_argument('--input_path', default='./temp/image.jpg', type=str, help='Directory of input images or path of single image')
parser.add_argument('--result_dir', default='./temp/', type=str, help='Directory for restored results')
parser.add_argument('--task', required=True, type=str, help='Task to run', choices=['Motion_Deblurring',
                                                                                    'Single_Image_Defocus_Deblurring',
                                                                                    'Deraining',
                                                                                    'Real_Denoising',
                                                                                    'Gaussian_Gray_Denoising',
                                                                                    'Gaussian_Color_Denoising'])

args = parser.parse_args()


def get_weights_and_parameters(task, parameters):
    if task == 'Motion_Deblurring':
        weights = os.path.join('Motion_Deblurring', 'pretrained_models', 'motion_deblurring.pth')
    elif task == 'Single_Image_Defocus_Deblurring':
        weights = os.path.join('Defocus_Deblurring', 'pretrained_models', 'single_image_defocus_deblurring.pth')
    elif task == 'Deraining':
        weights = os.path.join('Deraining', 'pretrained_models', 'deraining.pth')
    elif task == 'Real_Denoising':
        weights = os.path.join('Denoising', 'pretrained_models', 'real_denoising.pth')
        parameters['LayerNorm_type'] =  'BiasFree'
    return weights, parameters

task    = args.task
out_dir = os.path.join(args.result_dir, task)

os.makedirs(out_dir, exist_ok=True)

# Get model weights and parameters
parameters = {'inp_channels':3, 'out_channels':3, 'dim':48, 'num_blocks':[4,6,6,8], 'num_refinement_blocks':4, 'heads':[1,2,4,8], 'ffn_expansion_factor':2.66, 'bias':False, 'LayerNorm_type':'WithBias', 'dual_pixel_task':False}
weights, parameters = get_weights_and_parameters(task, parameters)

load_arch = run_path(os.path.join('basicsr', 'models', 'archs', 'restormer_arch.py'))
model = load_arch['Restormer'](**parameters)

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# device = torch.device('cpu')

model = model.to(device)
checkpoint = torch.load(weights)
model.load_state_dict(checkpoint['params'])

model.eval()


img_multiple_of = 8


with torch.inference_mode():

    img = cv2.cvtColor(cv2.imread(args.input_path), cv2.COLOR_BGR2RGB)

    input_ = torch.from_numpy(img).float().div(255.).permute(2,0,1).unsqueeze(0).to(device)

    # Pad the input if not_multiple_of 8
    h,w = input_.shape[2], input_.shape[3]
    H,W = ((h+img_multiple_of)//img_multiple_of)*img_multiple_of, ((w+img_multiple_of)//img_multiple_of)*img_multiple_of
    padh = H-h if h%img_multiple_of!=0 else 0
    padw = W-w if w%img_multiple_of!=0 else 0
    input_ = F.pad(input_, (0,padw,0,padh), 'reflect')
    
    restored = torch.clamp(model(input_),0,1)
    
    # Unpad the output
    restored = img_as_ubyte(restored[:,:,:h,:w].permute(0, 2, 3, 1).cpu().detach().numpy()[0])

    cv2.imwrite(os.path.join(out_dir, os.path.split(args.input_path)[-1]),cv2.cvtColor(restored, cv2.COLOR_RGB2BGR))