fix-1

SwinIR/infer.py

@@ -0,0 +1,93 @@
+import os
+
+import numpy as np
+import torch
+
+from SwinIR.models.network_swinir import SwinIR as net
+
+ROOT_PATH = os.path.dirname(__file__)
+
+
+class SwinIRDemo:
+    def __init__(self):
+        self.scale = 4
+        self.window_size = 8
+        self.tile = 800
+        self.tile_overlap = 32
+        self.device = 'cuda'
+        
+        model_path = os.path.join(ROOT_PATH, 'weight/003_realSR_BSRGAN_DFO_s64w8_SwinIR-M_x4_GAN.pth')
+        self.model = self.model_init(model_path)
+    
+    def model_init(self, model_path):
+        model = net(upscale=self.scale, in_chans=3, img_size=64, window_size=8,
+                    img_range=1., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='nearest+conv', resi_connection='1conv')
+        param_key_g = 'params_ema'
+        
+        pretrained_model = torch.load(model_path)
+        model.load_state_dict(
+                pretrained_model[param_key_g] if param_key_g in pretrained_model.keys() else pretrained_model,
+                strict=True)
+        
+        model.eval()
+        model = model.to(self.device)
+        return model
+    
+    def img_preprocess(self, img_PIL, device, window_size):
+        # imgname, img_lq, img_gt = get_image_pair(args, path)  # image to HWC-BGR, float32
+        # img_lq = cv2.imread(path, cv2.IMREAD_COLOR).astype(np.float32) / 255.
+        
+        # img_lq = img_PIL.convert('BGR')
+        img_lq = np.asarray(img_PIL)
+        img_lq = img_lq / 255
+        
+        img_lq = np.transpose(img_lq[:, :, [0, 1, 2]], (2, 0, 1))  # HCW-BGR to CHW-RGB
+        img_lq = torch.from_numpy(img_lq).float().unsqueeze(0).to(device)  # CHW-RGB to NCHW-RGB
+        
+        # pad input image to be a multiple of window_size
+        _, _, h_old, w_old = img_lq.size()
+        h_pad = (h_old // window_size + 1) * window_size - h_old
+        w_pad = (w_old // window_size + 1) * window_size - w_old
+        img_lq = torch.cat([img_lq, torch.flip(img_lq, [2])], 2)[:, :, :h_old + h_pad, :]
+        img_lq = torch.cat([img_lq, torch.flip(img_lq, [3])], 3)[:, :, :, :w_old + w_pad]
+        
+        return img_lq, h_old, w_old
+    
+    def test(self, img_lq):
+        b, c, h, w = img_lq.size()
+        tile = min(self.tile, h, w)
+        assert tile % self.window_size == 0, "tile size should be a multiple of window_size"
+        sf = self.scale
+        
+        stride = tile - self.tile_overlap
+        h_idx_list = list(range(0, h - tile, stride)) + [h - tile]
+        w_idx_list = list(range(0, w - tile, stride)) + [w - tile]
+        E = torch.zeros(b, c, h * sf, w * sf).type_as(img_lq)
+        W = torch.zeros_like(E)
+        
+        for h_idx in h_idx_list:
+            for w_idx in w_idx_list:
+                in_patch = img_lq[..., h_idx:h_idx + tile, w_idx:w_idx + tile]
+                out_patch = self.model(in_patch)
+                out_patch_mask = torch.ones_like(out_patch)
+                
+                E[..., h_idx * sf:(h_idx + tile) * sf, w_idx * sf:(w_idx + tile) * sf].add_(out_patch)
+                W[..., h_idx * sf:(h_idx + tile) * sf, w_idx * sf:(w_idx + tile) * sf].add_(out_patch_mask)
+        output = E.div_(W)
+        
+        return output
+    
+    def infer(self, img_lq):
+        img_lq, h_old, w_old = self.img_preprocess(img_lq, self.device, self.window_size)
+        
+        with torch.no_grad():
+            output = self.test(img_lq)
+            output = output[..., :h_old * self.scale, :w_old * self.scale]
+        
+        output = output.data.squeeze().float().cpu().clamp_(0, 1).numpy()
+        if output.ndim == 3:
+            output = np.transpose(output[[0, 1, 2], :, :], (1, 2, 0))  # CHW-RGB to HCW-BGR
+        output = (output * 255.0).round().astype(np.uint8)
+        
+        return output

SwinIR/main_test_swinir.py

@@ -0,0 +1,309 @@
+import argparse
+import cv2
+import glob
+import numpy as np
+from collections import OrderedDict
+import os
+import torch
+import requests
+
+from models.network_swinir import SwinIR as net
+from utils import util_calculate_psnr_ssim as util
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--task', type=str, default='color_dn', help='classical_sr, lightweight_sr, real_sr, '
+                                                                     'gray_dn, color_dn, jpeg_car, color_jpeg_car')
+    parser.add_argument('--scale', type=int, default=1, help='scale factor: 1, 2, 3, 4, 8') # 1 for dn and jpeg car
+    parser.add_argument('--noise', type=int, default=15, help='noise level: 15, 25, 50')
+    parser.add_argument('--jpeg', type=int, default=40, help='scale factor: 10, 20, 30, 40')
+    parser.add_argument('--training_patch_size', type=int, default=128, help='patch size used in training SwinIR. '
+                                       'Just used to differentiate two different settings in Table 2 of the paper. '
+                                       'Images are NOT tested patch by patch.')
+    parser.add_argument('--large_model', action='store_true', help='use large model, only provided for real image sr')
+    parser.add_argument('--model_path', type=str,
+                        default='model_zoo/swinir/001_classicalSR_DIV2K_s48w8_SwinIR-M_x2.pth')
+    parser.add_argument('--folder_lq', type=str, default=None, help='input low-quality test image folder')
+    parser.add_argument('--folder_gt', type=str, default=None, help='input ground-truth test image folder')
+    parser.add_argument('--tile', type=int, default=None, help='Tile size, None for no tile during testing (testing as a whole)')
+    parser.add_argument('--tile_overlap', type=int, default=32, help='Overlapping of different tiles')
+    args = parser.parse_args()
+
+    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    # set up model
+    if os.path.exists(args.model_path):
+        print(f'loading model from {args.model_path}')
+    else:
+        os.makedirs(os.path.dirname(args.model_path), exist_ok=True)
+        url = 'https://github.com/JingyunLiang/SwinIR/releases/download/v0.0/{}'.format(os.path.basename(args.model_path))
+        r = requests.get(url, allow_redirects=True)
+        print(f'downloading model {args.model_path}')
+        open(args.model_path, 'wb').write(r.content)
+
+    model = define_model(args)
+    model.eval()
+    model = model.to(device)
+
+    # setup folder and path
+    folder, save_dir, border, window_size = setup(args)
+    os.makedirs(save_dir, exist_ok=True)
+    test_results = OrderedDict()
+    test_results['psnr'] = []
+    test_results['ssim'] = []
+    test_results['psnr_y'] = []
+    test_results['ssim_y'] = []
+    test_results['psnrb'] = []
+    test_results['psnrb_y'] = []
+    psnr, ssim, psnr_y, ssim_y, psnrb, psnrb_y = 0, 0, 0, 0, 0, 0
+
+    for idx, path in enumerate(sorted(glob.glob(os.path.join(folder, '*')))):
+        # read image
+        imgname, img_lq, img_gt = get_image_pair(args, path)  # image to HWC-BGR, float32
+        img_lq = np.transpose(img_lq if img_lq.shape[2] == 1 else img_lq[:, :, [2, 1, 0]], (2, 0, 1))  # HCW-BGR to CHW-RGB
+        img_lq = torch.from_numpy(img_lq).float().unsqueeze(0).to(device)  # CHW-RGB to NCHW-RGB
+
+        # inference
+        with torch.no_grad():
+            # pad input image to be a multiple of window_size
+            _, _, h_old, w_old = img_lq.size()
+            h_pad = (h_old // window_size + 1) * window_size - h_old
+            w_pad = (w_old // window_size + 1) * window_size - w_old
+            img_lq = torch.cat([img_lq, torch.flip(img_lq, [2])], 2)[:, :, :h_old + h_pad, :]
+            img_lq = torch.cat([img_lq, torch.flip(img_lq, [3])], 3)[:, :, :, :w_old + w_pad]
+            output = test(img_lq, model, args, window_size)
+            output = output[..., :h_old * args.scale, :w_old * args.scale]
+
+        # save image
+        output = output.data.squeeze().float().cpu().clamp_(0, 1).numpy()
+        if output.ndim == 3:
+            output = np.transpose(output[[2, 1, 0], :, :], (1, 2, 0))  # CHW-RGB to HCW-BGR
+        output = (output * 255.0).round().astype(np.uint8)  # float32 to uint8
+        cv2.imwrite(f'{save_dir}/{imgname}_SwinIR.png', output)
+
+        # evaluate psnr/ssim/psnr_b
+        if img_gt is not None:
+            img_gt = (img_gt * 255.0).round().astype(np.uint8)  # float32 to uint8
+            img_gt = img_gt[:h_old * args.scale, :w_old * args.scale, ...]  # crop gt
+            img_gt = np.squeeze(img_gt)
+
+            psnr = util.calculate_psnr(output, img_gt, crop_border=border)
+            ssim = util.calculate_ssim(output, img_gt, crop_border=border)
+            test_results['psnr'].append(psnr)
+            test_results['ssim'].append(ssim)
+            if img_gt.ndim == 3:  # RGB image
+                psnr_y = util.calculate_psnr(output, img_gt, crop_border=border, test_y_channel=True)
+                ssim_y = util.calculate_ssim(output, img_gt, crop_border=border, test_y_channel=True)
+                test_results['psnr_y'].append(psnr_y)
+                test_results['ssim_y'].append(ssim_y)
+            if args.task in ['jpeg_car', 'color_jpeg_car']:
+                psnrb = util.calculate_psnrb(output, img_gt, crop_border=border, test_y_channel=False)
+                test_results['psnrb'].append(psnrb)
+                if args.task in ['color_jpeg_car']:
+                    psnrb_y = util.calculate_psnrb(output, img_gt, crop_border=border, test_y_channel=True)
+                    test_results['psnrb_y'].append(psnrb_y)
+            print('Testing {:d} {:20s} - PSNR: {:.2f} dB; SSIM: {:.4f}; PSNRB: {:.2f} dB;'
+                  'PSNR_Y: {:.2f} dB; SSIM_Y: {:.4f}; PSNRB_Y: {:.2f} dB.'.
+                  format(idx, imgname, psnr, ssim, psnrb, psnr_y, ssim_y, psnrb_y))
+        else:
+            print('Testing {:d} {:20s}'.format(idx, imgname))
+
+    # summarize psnr/ssim
+    if img_gt is not None:
+        ave_psnr = sum(test_results['psnr']) / len(test_results['psnr'])
+        ave_ssim = sum(test_results['ssim']) / len(test_results['ssim'])
+        print('\n{} \n-- Average PSNR/SSIM(RGB): {:.2f} dB; {:.4f}'.format(save_dir, ave_psnr, ave_ssim))
+        if img_gt.ndim == 3:
+            ave_psnr_y = sum(test_results['psnr_y']) / len(test_results['psnr_y'])
+            ave_ssim_y = sum(test_results['ssim_y']) / len(test_results['ssim_y'])
+            print('-- Average PSNR_Y/SSIM_Y: {:.2f} dB; {:.4f}'.format(ave_psnr_y, ave_ssim_y))
+        if args.task in ['jpeg_car', 'color_jpeg_car']:
+            ave_psnrb = sum(test_results['psnrb']) / len(test_results['psnrb'])
+            print('-- Average PSNRB: {:.2f} dB'.format(ave_psnrb))
+            if args.task in ['color_jpeg_car']:
+                ave_psnrb_y = sum(test_results['psnrb_y']) / len(test_results['psnrb_y'])
+                print('-- Average PSNRB_Y: {:.2f} dB'.format(ave_psnrb_y))
+
+
+def define_model(args):
+    # 001 classical image sr
+    if args.task == 'classical_sr':
+        model = net(upscale=args.scale, in_chans=3, img_size=args.training_patch_size, window_size=8,
+                    img_range=1., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='pixelshuffle', resi_connection='1conv')
+        param_key_g = 'params'
+
+    # 002 lightweight image sr
+    # use 'pixelshuffledirect' to save parameters
+    elif args.task == 'lightweight_sr':
+        model = net(upscale=args.scale, in_chans=3, img_size=64, window_size=8,
+                    img_range=1., depths=[6, 6, 6, 6], embed_dim=60, num_heads=[6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='pixelshuffledirect', resi_connection='1conv')
+        param_key_g = 'params'
+
+    # 003 real-world image sr
+    elif args.task == 'real_sr':
+        if not args.large_model:
+            # use 'nearest+conv' to avoid block artifacts
+            model = net(upscale=args.scale, in_chans=3, img_size=64, window_size=8,
+                        img_range=1., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                        mlp_ratio=2, upsampler='nearest+conv', resi_connection='1conv')
+        else:
+            # larger model size; use '3conv' to save parameters and memory; use ema for GAN training
+            model = net(upscale=args.scale, in_chans=3, img_size=64, window_size=8,
+                        img_range=1., depths=[6, 6, 6, 6, 6, 6, 6, 6, 6], embed_dim=240,
+                        num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8],
+                        mlp_ratio=2, upsampler='nearest+conv', resi_connection='3conv')
+        param_key_g = 'params_ema'
+
+    # 004 grayscale image denoising
+    elif args.task == 'gray_dn':
+        model = net(upscale=1, in_chans=1, img_size=128, window_size=8,
+                    img_range=1., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='', resi_connection='1conv')
+        param_key_g = 'params'
+
+    # 005 color image denoising
+    elif args.task == 'color_dn':
+        model = net(upscale=1, in_chans=3, img_size=128, window_size=8,
+                    img_range=1., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='', resi_connection='1conv')
+        param_key_g = 'params'
+
+    # 006 grayscale JPEG compression artifact reduction
+    # use window_size=7 because JPEG encoding uses 8x8; use img_range=255 because it's sligtly better than 1
+    elif args.task == 'jpeg_car':
+        model = net(upscale=1, in_chans=1, img_size=126, window_size=7,
+                    img_range=255., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='', resi_connection='1conv')
+        param_key_g = 'params'
+
+    # 006 color JPEG compression artifact reduction
+    # use window_size=7 because JPEG encoding uses 8x8; use img_range=255 because it's sligtly better than 1
+    elif args.task == 'color_jpeg_car':
+        model = net(upscale=1, in_chans=3, img_size=126, window_size=7,
+                    img_range=255., depths=[6, 6, 6, 6, 6, 6], embed_dim=180, num_heads=[6, 6, 6, 6, 6, 6],
+                    mlp_ratio=2, upsampler='', resi_connection='1conv')
+        param_key_g = 'params'
+
+    pretrained_model = torch.load(args.model_path)
+    model.load_state_dict(pretrained_model[param_key_g] if param_key_g in pretrained_model.keys() else pretrained_model, strict=True)
+
+    return model
+
+
+def setup(args):
+    # 001 classical image sr/ 002 lightweight image sr
+    if args.task in ['classical_sr', 'lightweight_sr']:
+        save_dir = f'results/swinir_{args.task}_x{args.scale}'
+        folder = args.folder_gt
+        border = args.scale
+        window_size = 8
+
+    # 003 real-world image sr
+    elif args.task in ['real_sr']:
+        save_dir = f'results/swinir_{args.task}_x{args.scale}'
+        if args.large_model:
+            save_dir += '_large'
+        folder = args.folder_lq
+        border = 0
+        window_size = 8
+
+    # 004 grayscale image denoising/ 005 color image denoising
+    elif args.task in ['gray_dn', 'color_dn']:
+        save_dir = f'results/swinir_{args.task}_noise{args.noise}'
+        folder = args.folder_gt
+        border = 0
+        window_size = 8
+
+    # 006 JPEG compression artifact reduction
+    elif args.task in ['jpeg_car', 'color_jpeg_car']:
+        save_dir = f'results/swinir_{args.task}_jpeg{args.jpeg}'
+        folder = args.folder_gt
+        border = 0
+        window_size = 7
+
+    return folder, save_dir, border, window_size
+
+
+def get_image_pair(args, path):
+    (imgname, imgext) = os.path.splitext(os.path.basename(path))
+
+    # 001 classical image sr/ 002 lightweight image sr (load lq-gt image pairs)
+    if args.task in ['classical_sr', 'lightweight_sr']:
+        img_gt = cv2.imread(path, cv2.IMREAD_COLOR).astype(np.float32) / 255.
+        img_lq = cv2.imread(f'{args.folder_lq}/{imgname}x{args.scale}{imgext}', cv2.IMREAD_COLOR).astype(
+            np.float32) / 255.
+
+    # 003 real-world image sr (load lq image only)
+    elif args.task in ['real_sr']:
+        img_gt = None
+        img_lq = cv2.imread(path, cv2.IMREAD_COLOR).astype(np.float32) / 255.
+
+    # 004 grayscale image denoising (load gt image and generate lq image on-the-fly)
+    elif args.task in ['gray_dn']:
+        img_gt = cv2.imread(path, cv2.IMREAD_GRAYSCALE).astype(np.float32) / 255.
+        np.random.seed(seed=0)
+        img_lq = img_gt + np.random.normal(0, args.noise / 255., img_gt.shape)
+        img_gt = np.expand_dims(img_gt, axis=2)
+        img_lq = np.expand_dims(img_lq, axis=2)
+
+    # 005 color image denoising (load gt image and generate lq image on-the-fly)
+    elif args.task in ['color_dn']:
+        img_gt = cv2.imread(path, cv2.IMREAD_COLOR).astype(np.float32) / 255.
+        np.random.seed(seed=0)
+        img_lq = img_gt + np.random.normal(0, args.noise / 255., img_gt.shape)
+
+    # 006 grayscale JPEG compression artifact reduction (load gt image and generate lq image on-the-fly)
+    elif args.task in ['jpeg_car']:
+        img_gt = cv2.imread(path, cv2.IMREAD_UNCHANGED)
+        if img_gt.ndim != 2:
+            img_gt = util.bgr2ycbcr(img_gt, y_only=True)
+        result, encimg = cv2.imencode('.jpg', img_gt, [int(cv2.IMWRITE_JPEG_QUALITY), args.jpeg])
+        img_lq = cv2.imdecode(encimg, 0)
+        img_gt = np.expand_dims(img_gt, axis=2).astype(np.float32) / 255.
+        img_lq = np.expand_dims(img_lq, axis=2).astype(np.float32) / 255.
+
+    # 006 JPEG compression artifact reduction (load gt image and generate lq image on-the-fly)
+    elif args.task in ['color_jpeg_car']:
+        img_gt = cv2.imread(path)
+        result, encimg = cv2.imencode('.jpg', img_gt, [int(cv2.IMWRITE_JPEG_QUALITY), args.jpeg])
+        img_lq = cv2.imdecode(encimg, 1)
+        img_gt = img_gt.astype(np.float32)/ 255.
+        img_lq = img_lq.astype(np.float32)/ 255.
+
+    return imgname, img_lq, img_gt
+
+
+def test(img_lq, model, args, window_size):
+    if args.tile is None:
+        # test the image as a whole
+        output = model(img_lq)
+    else:
+        # test the image tile by tile
+        b, c, h, w = img_lq.size()
+        tile = min(args.tile, h, w)
+        assert tile % window_size == 0, "tile size should be a multiple of window_size"
+        tile_overlap = args.tile_overlap
+        sf = args.scale
+
+        stride = tile - tile_overlap
+        h_idx_list = list(range(0, h-tile, stride)) + [h-tile]
+        w_idx_list = list(range(0, w-tile, stride)) + [w-tile]
+        E = torch.zeros(b, c, h*sf, w*sf).type_as(img_lq)
+        W = torch.zeros_like(E)
+
+        for h_idx in h_idx_list:
+            for w_idx in w_idx_list:
+                in_patch = img_lq[..., h_idx:h_idx+tile, w_idx:w_idx+tile]
+                out_patch = model(in_patch)
+                out_patch_mask = torch.ones_like(out_patch)
+
+                E[..., h_idx*sf:(h_idx+tile)*sf, w_idx*sf:(w_idx+tile)*sf].add_(out_patch)
+                W[..., h_idx*sf:(h_idx+tile)*sf, w_idx*sf:(w_idx+tile)*sf].add_(out_patch_mask)
+        output = E.div_(W)
+
+    return output
+
+if __name__ == '__main__':
+    main()

SwinIR/models/network_swinir.py

@@ -0,0 +1,867 @@
+# -----------------------------------------------------------------------------------
+# SwinIR: Image Restoration Using Swin Transformer, https://arxiv.org/abs/2108.10257
+# Originally Written by Ze Liu, Modified by Jingyun Liang.
+# -----------------------------------------------------------------------------------
+
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.checkpoint as checkpoint
+from timm.models.layers import DropPath, to_2tuple, trunc_normal_
+
+
+class Mlp(nn.Module):
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+def window_partition(x, window_size):
+    """
+    Args:
+        x: (B, H, W, C)
+        window_size (int): window size
+
+    Returns:
+        windows: (num_windows*B, window_size, window_size, C)
+    """
+    B, H, W, C = x.shape
+    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows
+
+
+def window_reverse(windows, window_size, H, W):
+    """
+    Args:
+        windows: (num_windows*B, window_size, window_size, C)
+        window_size (int): Window size
+        H (int): Height of image
+        W (int): Width of image
+
+    Returns:
+        x: (B, H, W, C)
+    """
+    B = int(windows.shape[0] / (H * W / window_size / window_size))
+    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
+    return x
+
+
+class WindowAttention(nn.Module):
+    r""" Window based multi-head self attention (W-MSA) module with relative position bias.
+    It supports both of shifted and non-shifted window.
+
+    Args:
+        dim (int): Number of input channels.
+        window_size (tuple[int]): The height and width of the window.
+        num_heads (int): Number of attention heads.
+        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
+        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
+        proj_drop (float, optional): Dropout ratio of output. Default: 0.0
+    """
+
+    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
+
+        super().__init__()
+        self.dim = dim
+        self.window_size = window_size  # Wh, Ww
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim ** -0.5
+
+        # define a parameter table of relative position bias
+        self.relative_position_bias_table = nn.Parameter(
+            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))  # 2*Wh-1 * 2*Ww-1, nH
+
+        # get pair-wise relative position index for each token inside the window
+        coords_h = torch.arange(self.window_size[0])
+        coords_w = torch.arange(self.window_size[1])
+        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
+        relative_coords[:, :, 1] += self.window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
+        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+        self.register_buffer("relative_position_index", relative_position_index)
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+
+        self.proj_drop = nn.Dropout(proj_drop)
+
+        trunc_normal_(self.relative_position_bias_table, std=.02)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x, mask=None):
+        """
+        Args:
+            x: input features with shape of (num_windows*B, N, C)
+            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
+        """
+        B_, N, C = x.shape
+        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)
+
+        q = q * self.scale
+        attn = (q @ k.transpose(-2, -1))
+
+        relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
+            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)  # Wh*Ww,Wh*Ww,nH
+        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
+        attn = attn + relative_position_bias.unsqueeze(0)
+
+        if mask is not None:
+            nW = mask.shape[0]
+            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
+            attn = attn.view(-1, self.num_heads, N, N)
+            attn = self.softmax(attn)
+        else:
+            attn = self.softmax(attn)
+
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+    def extra_repr(self) -> str:
+        return f'dim={self.dim}, window_size={self.window_size}, num_heads={self.num_heads}'
+
+    def flops(self, N):
+        # calculate flops for 1 window with token length of N
+        flops = 0
+        # qkv = self.qkv(x)
+        flops += N * self.dim * 3 * self.dim
+        # attn = (q @ k.transpose(-2, -1))
+        flops += self.num_heads * N * (self.dim // self.num_heads) * N
+        #  x = (attn @ v)
+        flops += self.num_heads * N * N * (self.dim // self.num_heads)
+        # x = self.proj(x)
+        flops += N * self.dim * self.dim
+        return flops
+
+
+class SwinTransformerBlock(nn.Module):
+    r""" Swin Transformer Block.
+
+    Args:
+        dim (int): Number of input channels.
+        input_resolution (tuple[int]): Input resulotion.
+        num_heads (int): Number of attention heads.
+        window_size (int): Window size.
+        shift_size (int): Shift size for SW-MSA.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float, optional): Stochastic depth rate. Default: 0.0
+        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,
+                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
+                 act_layer=nn.GELU, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.dim = dim
+        self.input_resolution = input_resolution
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.shift_size = shift_size
+        self.mlp_ratio = mlp_ratio
+        if min(self.input_resolution) <= self.window_size:
+            # if window size is larger than input resolution, we don't partition windows
+            self.shift_size = 0
+            self.window_size = min(self.input_resolution)
+        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
+
+        self.norm1 = norm_layer(dim)
+        self.attn = WindowAttention(
+            dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,
+            qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
+
+        if self.shift_size > 0:
+            attn_mask = self.calculate_mask(self.input_resolution)
+        else:
+            attn_mask = None
+
+        self.register_buffer("attn_mask", attn_mask)
+
+    def calculate_mask(self, x_size):
+        # calculate attention mask for SW-MSA
+        H, W = x_size
+        img_mask = torch.zeros((1, H, W, 1))  # 1 H W 1
+        h_slices = (slice(0, -self.window_size),
+                    slice(-self.window_size, -self.shift_size),
+                    slice(-self.shift_size, None))
+        w_slices = (slice(0, -self.window_size),
+                    slice(-self.window_size, -self.shift_size),
+                    slice(-self.shift_size, None))
+        cnt = 0
+        for h in h_slices:
+            for w in w_slices:
+                img_mask[:, h, w, :] = cnt
+                cnt += 1
+
+        mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1
+        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
+        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
+
+        return attn_mask
+
+    def forward(self, x, x_size):
+        H, W = x_size
+        B, L, C = x.shape
+        # assert L == H * W, "input feature has wrong size"
+
+        shortcut = x
+        x = self.norm1(x)
+        x = x.view(B, H, W, C)
+
+        # cyclic shift
+        if self.shift_size > 0:
+            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
+        else:
+            shifted_x = x
+
+        # partition windows
+        x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
+        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # nW*B, window_size*window_size, C
+
+        # W-MSA/SW-MSA (to be compatible for testing on images whose shapes are the multiple of window size
+        if self.input_resolution == x_size:
+            attn_windows = self.attn(x_windows, mask=self.attn_mask)  # nW*B, window_size*window_size, C
+        else:
+            attn_windows = self.attn(x_windows, mask=self.calculate_mask(x_size).to(x.device))
+
+        # merge windows
+        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
+        shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C
+
+        # reverse cyclic shift
+        if self.shift_size > 0:
+            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
+        else:
+            x = shifted_x
+        x = x.view(B, H * W, C)
+
+        # FFN
+        x = shortcut + self.drop_path(x)
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+
+        return x
+
+    def extra_repr(self) -> str:
+        return f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " \
+               f"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}"
+
+    def flops(self):
+        flops = 0
+        H, W = self.input_resolution
+        # norm1
+        flops += self.dim * H * W
+        # W-MSA/SW-MSA
+        nW = H * W / self.window_size / self.window_size
+        flops += nW * self.attn.flops(self.window_size * self.window_size)
+        # mlp
+        flops += 2 * H * W * self.dim * self.dim * self.mlp_ratio
+        # norm2
+        flops += self.dim * H * W
+        return flops
+
+
+class PatchMerging(nn.Module):
+    r""" Patch Merging Layer.
+
+    Args:
+        input_resolution (tuple[int]): Resolution of input feature.
+        dim (int): Number of input channels.
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.input_resolution = input_resolution
+        self.dim = dim
+        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
+        self.norm = norm_layer(4 * dim)
+
+    def forward(self, x):
+        """
+        x: B, H*W, C
+        """
+        H, W = self.input_resolution
+        B, L, C = x.shape
+        assert L == H * W, "input feature has wrong size"
+        assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."
+
+        x = x.view(B, H, W, C)
+
+        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
+        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
+        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
+        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
+        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
+        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
+
+        x = self.norm(x)
+        x = self.reduction(x)
+
+        return x
+
+    def extra_repr(self) -> str:
+        return f"input_resolution={self.input_resolution}, dim={self.dim}"
+
+    def flops(self):
+        H, W = self.input_resolution
+        flops = H * W * self.dim
+        flops += (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim
+        return flops
+
+
+class BasicLayer(nn.Module):
+    """ A basic Swin Transformer layer for one stage.
+
+    Args:
+        dim (int): Number of input channels.
+        input_resolution (tuple[int]): Input resolution.
+        depth (int): Number of blocks.
+        num_heads (int): Number of attention heads.
+        window_size (int): Local window size.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
+        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
+        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+    """
+
+    def __init__(self, dim, input_resolution, depth, num_heads, window_size,
+                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False):
+
+        super().__init__()
+        self.dim = dim
+        self.input_resolution = input_resolution
+        self.depth = depth
+        self.use_checkpoint = use_checkpoint
+
+        # build blocks
+        self.blocks = nn.ModuleList([
+            SwinTransformerBlock(dim=dim, input_resolution=input_resolution,
+                                 num_heads=num_heads, window_size=window_size,
+                                 shift_size=0 if (i % 2 == 0) else window_size // 2,
+                                 mlp_ratio=mlp_ratio,
+                                 qkv_bias=qkv_bias, qk_scale=qk_scale,
+                                 drop=drop, attn_drop=attn_drop,
+                                 drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
+                                 norm_layer=norm_layer)
+            for i in range(depth)])
+
+        # patch merging layer
+        if downsample is not None:
+            self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
+        else:
+            self.downsample = None
+
+    def forward(self, x, x_size):
+        for blk in self.blocks:
+            if self.use_checkpoint:
+                x = checkpoint.checkpoint(blk, x, x_size)
+            else:
+                x = blk(x, x_size)
+        if self.downsample is not None:
+            x = self.downsample(x)
+        return x
+
+    def extra_repr(self) -> str:
+        return f"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}"
+
+    def flops(self):
+        flops = 0
+        for blk in self.blocks:
+            flops += blk.flops()
+        if self.downsample is not None:
+            flops += self.downsample.flops()
+        return flops
+
+
+class RSTB(nn.Module):
+    """Residual Swin Transformer Block (RSTB).
+
+    Args:
+        dim (int): Number of input channels.
+        input_resolution (tuple[int]): Input resolution.
+        depth (int): Number of blocks.
+        num_heads (int): Number of attention heads.
+        window_size (int): Local window size.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
+        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
+        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+        img_size: Input image size.
+        patch_size: Patch size.
+        resi_connection: The convolutional block before residual connection.
+    """
+
+    def __init__(self, dim, input_resolution, depth, num_heads, window_size,
+                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False,
+                 img_size=224, patch_size=4, resi_connection='1conv'):
+        super(RSTB, self).__init__()
+
+        self.dim = dim
+        self.input_resolution = input_resolution
+
+        self.residual_group = BasicLayer(dim=dim,
+                                         input_resolution=input_resolution,
+                                         depth=depth,
+                                         num_heads=num_heads,
+                                         window_size=window_size,
+                                         mlp_ratio=mlp_ratio,
+                                         qkv_bias=qkv_bias, qk_scale=qk_scale,
+                                         drop=drop, attn_drop=attn_drop,
+                                         drop_path=drop_path,
+                                         norm_layer=norm_layer,
+                                         downsample=downsample,
+                                         use_checkpoint=use_checkpoint)
+
+        if resi_connection == '1conv':
+            self.conv = nn.Conv2d(dim, dim, 3, 1, 1)
+        elif resi_connection == '3conv':
+            # to save parameters and memory
+            self.conv = nn.Sequential(nn.Conv2d(dim, dim // 4, 3, 1, 1), nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                                      nn.Conv2d(dim // 4, dim // 4, 1, 1, 0),
+                                      nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                                      nn.Conv2d(dim // 4, dim, 3, 1, 1))
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=0, embed_dim=dim,
+            norm_layer=None)
+
+        self.patch_unembed = PatchUnEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=0, embed_dim=dim,
+            norm_layer=None)
+
+    def forward(self, x, x_size):
+        return self.patch_embed(self.conv(self.patch_unembed(self.residual_group(x, x_size), x_size))) + x
+
+    def flops(self):
+        flops = 0
+        flops += self.residual_group.flops()
+        H, W = self.input_resolution
+        flops += H * W * self.dim * self.dim * 9
+        flops += self.patch_embed.flops()
+        flops += self.patch_unembed.flops()
+
+        return flops
+
+
+class PatchEmbed(nn.Module):
+    r""" Image to Patch Embedding
+
+    Args:
+        img_size (int): Image size.  Default: 224.
+        patch_size (int): Patch token size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        norm_layer (nn.Module, optional): Normalization layer. Default: None
+    """
+
+    def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.patches_resolution = patches_resolution
+        self.num_patches = patches_resolution[0] * patches_resolution[1]
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        if norm_layer is not None:
+            self.norm = norm_layer(embed_dim)
+        else:
+            self.norm = None
+
+    def forward(self, x):
+        x = x.flatten(2).transpose(1, 2)  # B Ph*Pw C
+        if self.norm is not None:
+            x = self.norm(x)
+        return x
+
+    def flops(self):
+        flops = 0
+        H, W = self.img_size
+        if self.norm is not None:
+            flops += H * W * self.embed_dim
+        return flops
+
+
+class PatchUnEmbed(nn.Module):
+    r""" Image to Patch Unembedding
+
+    Args:
+        img_size (int): Image size.  Default: 224.
+        patch_size (int): Patch token size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        norm_layer (nn.Module, optional): Normalization layer. Default: None
+    """
+
+    def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.patches_resolution = patches_resolution
+        self.num_patches = patches_resolution[0] * patches_resolution[1]
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+    def forward(self, x, x_size):
+        B, HW, C = x.shape
+        x = x.transpose(1, 2).view(B, self.embed_dim, x_size[0], x_size[1])  # B Ph*Pw C
+        return x
+
+    def flops(self):
+        flops = 0
+        return flops
+
+
+class Upsample(nn.Sequential):
+    """Upsample module.
+
+    Args:
+        scale (int): Scale factor. Supported scales: 2^n and 3.
+        num_feat (int): Channel number of intermediate features.
+    """
+
+    def __init__(self, scale, num_feat):
+        m = []
+        if (scale & (scale - 1)) == 0:  # scale = 2^n
+            for _ in range(int(math.log(scale, 2))):
+                m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1))
+                m.append(nn.PixelShuffle(2))
+        elif scale == 3:
+            m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1))
+            m.append(nn.PixelShuffle(3))
+        else:
+            raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.')
+        super(Upsample, self).__init__(*m)
+
+
+class UpsampleOneStep(nn.Sequential):
+    """UpsampleOneStep module (the difference with Upsample is that it always only has 1conv + 1pixelshuffle)
+       Used in lightweight SR to save parameters.
+
+    Args:
+        scale (int): Scale factor. Supported scales: 2^n and 3.
+        num_feat (int): Channel number of intermediate features.
+
+    """
+
+    def __init__(self, scale, num_feat, num_out_ch, input_resolution=None):
+        self.num_feat = num_feat
+        self.input_resolution = input_resolution
+        m = []
+        m.append(nn.Conv2d(num_feat, (scale ** 2) * num_out_ch, 3, 1, 1))
+        m.append(nn.PixelShuffle(scale))
+        super(UpsampleOneStep, self).__init__(*m)
+
+    def flops(self):
+        H, W = self.input_resolution
+        flops = H * W * self.num_feat * 3 * 9
+        return flops
+
+
+class SwinIR(nn.Module):
+    r""" SwinIR
+        A PyTorch impl of : `SwinIR: Image Restoration Using Swin Transformer`, based on Swin Transformer.
+
+    Args:
+        img_size (int | tuple(int)): Input image size. Default 64
+        patch_size (int | tuple(int)): Patch size. Default: 1
+        in_chans (int): Number of input image channels. Default: 3
+        embed_dim (int): Patch embedding dimension. Default: 96
+        depths (tuple(int)): Depth of each Swin Transformer layer.
+        num_heads (tuple(int)): Number of attention heads in different layers.
+        window_size (int): Window size. Default: 7
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4
+        qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float): Override default qk scale of head_dim ** -0.5 if set. Default: None
+        drop_rate (float): Dropout rate. Default: 0
+        attn_drop_rate (float): Attention dropout rate. Default: 0
+        drop_path_rate (float): Stochastic depth rate. Default: 0.1
+        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
+        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False
+        patch_norm (bool): If True, add normalization after patch embedding. Default: True
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False
+        upscale: Upscale factor. 2/3/4/8 for image SR, 1 for denoising and compress artifact reduction
+        img_range: Image range. 1. or 255.
+        upsampler: The reconstruction reconstruction module. 'pixelshuffle'/'pixelshuffledirect'/'nearest+conv'/None
+        resi_connection: The convolutional block before residual connection. '1conv'/'3conv'
+    """
+
+    def __init__(self, img_size=64, patch_size=1, in_chans=3,
+                 embed_dim=96, depths=[6, 6, 6, 6], num_heads=[6, 6, 6, 6],
+                 window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None,
+                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,
+                 norm_layer=nn.LayerNorm, ape=False, patch_norm=True,
+                 use_checkpoint=False, upscale=2, img_range=1., upsampler='', resi_connection='1conv',
+                 **kwargs):
+        super(SwinIR, self).__init__()
+        num_in_ch = in_chans
+        num_out_ch = in_chans
+        num_feat = 64
+        self.img_range = img_range
+        if in_chans == 3:
+            rgb_mean = (0.4488, 0.4371, 0.4040)
+            self.mean = torch.Tensor(rgb_mean).view(1, 3, 1, 1)
+        else:
+            self.mean = torch.zeros(1, 1, 1, 1)
+        self.upscale = upscale
+        self.upsampler = upsampler
+        self.window_size = window_size
+
+        #####################################################################################################
+        ################################### 1, shallow feature extraction ###################################
+        self.conv_first = nn.Conv2d(num_in_ch, embed_dim, 3, 1, 1)
+
+        #####################################################################################################
+        ################################### 2, deep feature extraction ######################################
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.ape = ape
+        self.patch_norm = patch_norm
+        self.num_features = embed_dim
+        self.mlp_ratio = mlp_ratio
+
+        # split image into non-overlapping patches
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=embed_dim, embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None)
+        num_patches = self.patch_embed.num_patches
+        patches_resolution = self.patch_embed.patches_resolution
+        self.patches_resolution = patches_resolution
+
+        # merge non-overlapping patches into image
+        self.patch_unembed = PatchUnEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=embed_dim, embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None)
+
+        # absolute position embedding
+        if self.ape:
+            self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
+            trunc_normal_(self.absolute_pos_embed, std=.02)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        # stochastic depth
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule
+
+        # build Residual Swin Transformer blocks (RSTB)
+        self.layers = nn.ModuleList()
+        for i_layer in range(self.num_layers):
+            layer = RSTB(dim=embed_dim,
+                         input_resolution=(patches_resolution[0],
+                                           patches_resolution[1]),
+                         depth=depths[i_layer],
+                         num_heads=num_heads[i_layer],
+                         window_size=window_size,
+                         mlp_ratio=self.mlp_ratio,
+                         qkv_bias=qkv_bias, qk_scale=qk_scale,
+                         drop=drop_rate, attn_drop=attn_drop_rate,
+                         drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],  # no impact on SR results
+                         norm_layer=norm_layer,
+                         downsample=None,
+                         use_checkpoint=use_checkpoint,
+                         img_size=img_size,
+                         patch_size=patch_size,
+                         resi_connection=resi_connection
+
+                         )
+            self.layers.append(layer)
+        self.norm = norm_layer(self.num_features)
+
+        # build the last conv layer in deep feature extraction
+        if resi_connection == '1conv':
+            self.conv_after_body = nn.Conv2d(embed_dim, embed_dim, 3, 1, 1)
+        elif resi_connection == '3conv':
+            # to save parameters and memory
+            self.conv_after_body = nn.Sequential(nn.Conv2d(embed_dim, embed_dim // 4, 3, 1, 1),
+                                                 nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                                                 nn.Conv2d(embed_dim // 4, embed_dim // 4, 1, 1, 0),
+                                                 nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                                                 nn.Conv2d(embed_dim // 4, embed_dim, 3, 1, 1))
+
+        #####################################################################################################
+        ################################ 3, high quality image reconstruction ################################
+        if self.upsampler == 'pixelshuffle':
+            # for classical SR
+            self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1),
+                                                      nn.LeakyReLU(inplace=True))
+            self.upsample = Upsample(upscale, num_feat)
+            self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1)
+        elif self.upsampler == 'pixelshuffledirect':
+            # for lightweight SR (to save parameters)
+            self.upsample = UpsampleOneStep(upscale, embed_dim, num_out_ch,
+                                            (patches_resolution[0], patches_resolution[1]))
+        elif self.upsampler == 'nearest+conv':
+            # for real-world SR (less artifacts)
+            self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1),
+                                                      nn.LeakyReLU(inplace=True))
+            self.conv_up1 = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+            if self.upscale == 4:
+                self.conv_up2 = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+            self.conv_hr = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+            self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1)
+            self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
+        else:
+            # for image denoising and JPEG compression artifact reduction
+            self.conv_last = nn.Conv2d(embed_dim, num_out_ch, 3, 1, 1)
+
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'absolute_pos_embed'}
+
+    @torch.jit.ignore
+    def no_weight_decay_keywords(self):
+        return {'relative_position_bias_table'}
+
+    def check_image_size(self, x):
+        _, _, h, w = x.size()
+        mod_pad_h = (self.window_size - h % self.window_size) % self.window_size
+        mod_pad_w = (self.window_size - w % self.window_size) % self.window_size
+        x = F.pad(x, (0, mod_pad_w, 0, mod_pad_h), 'reflect')
+        return x
+
+    def forward_features(self, x):
+        x_size = (x.shape[2], x.shape[3])
+        x = self.patch_embed(x)
+        if self.ape:
+            x = x + self.absolute_pos_embed
+        x = self.pos_drop(x)
+
+        for layer in self.layers:
+            x = layer(x, x_size)
+
+        x = self.norm(x)  # B L C
+        x = self.patch_unembed(x, x_size)
+
+        return x
+
+    def forward(self, x):
+        H, W = x.shape[2:]
+        x = self.check_image_size(x)
+
+        self.mean = self.mean.type_as(x)
+        x = (x - self.mean) * self.img_range
+
+        if self.upsampler == 'pixelshuffle':
+            # for classical SR
+            x = self.conv_first(x)
+            x = self.conv_after_body(self.forward_features(x)) + x
+            x = self.conv_before_upsample(x)
+            x = self.conv_last(self.upsample(x))
+        elif self.upsampler == 'pixelshuffledirect':
+            # for lightweight SR
+            x = self.conv_first(x)
+            x = self.conv_after_body(self.forward_features(x)) + x
+            x = self.upsample(x)
+        elif self.upsampler == 'nearest+conv':
+            # for real-world SR
+            x = self.conv_first(x)
+            x = self.conv_after_body(self.forward_features(x)) + x
+            x = self.conv_before_upsample(x)
+            x = self.lrelu(self.conv_up1(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest')))
+            if self.upscale == 4:
+                x = self.lrelu(self.conv_up2(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest')))
+            x = self.conv_last(self.lrelu(self.conv_hr(x)))
+        else:
+            # for image denoising and JPEG compression artifact reduction
+            x_first = self.conv_first(x)
+            res = self.conv_after_body(self.forward_features(x_first)) + x_first
+            x = x + self.conv_last(res)
+
+        x = x / self.img_range + self.mean
+
+        return x[:, :, :H * self.upscale, :W * self.upscale]
+
+    def flops(self):
+        flops = 0
+        H, W = self.patches_resolution
+        flops += H * W * 3 * self.embed_dim * 9
+        flops += self.patch_embed.flops()
+        for i, layer in enumerate(self.layers):
+            flops += layer.flops()
+        flops += H * W * 3 * self.embed_dim * self.embed_dim
+        flops += self.upsample.flops()
+        return flops
+
+
+if __name__ == '__main__':
+    upscale = 4
+    window_size = 8
+    height = (1024 // upscale // window_size + 1) * window_size
+    width = (720 // upscale // window_size + 1) * window_size
+    model = SwinIR(upscale=2, img_size=(height, width),
+                   window_size=window_size, img_range=1., depths=[6, 6, 6, 6],
+                   embed_dim=60, num_heads=[6, 6, 6, 6], mlp_ratio=2, upsampler='pixelshuffledirect')
+    print(model)
+    print(height, width, model.flops() / 1e9)
+
+    x = torch.randn((1, 3, height, width))
+    x = model(x)
+    print(x.shape)

SwinIR/weight/003_realSR_BSRGAN_DFO_s64w8_SwinIR-M_x4_GAN.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9afb61e65e04eb7f8aba5095d070bbe9af28df76acd0c9405aeb33b814bcfc6
+size 67129861

app.py

@@ -1,109 +1,40 @@
-import importlib
-from collections import OrderedDict
-from pathlib import Path
-
+import gradio
 import gradio as gr
 import os
 
-import numpy as np
-import torch
-from PIL import Image
-from torchvision import transforms
-
-from sam_diffsr.utils_sr.hparams import set_hparams, hparams
-from sam_diffsr.utils_sr.matlab_resize import imresize
-
-
-def get_img_data(img_PIL, hparams, sr_scale=4):
-    img_lr = img_PIL.convert('RGB')
-    img_lr = np.uint8(np.asarray(img_lr))
-
-    h, w, c = img_lr.shape
-    h, w = h * sr_scale, w * sr_scale
-    h = h - h % (sr_scale * 2)
-    w = w - w % (sr_scale * 2)
-    h_l = h // sr_scale
-    w_l = w // sr_scale
-
-    img_lr = img_lr[:h_l, :w_l]
-
-    to_tensor_norm = transforms.Compose([
-        transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
-    ])
-
-    img_lr_up = imresize(img_lr / 256, hparams['sr_scale'])  # np.float [H, W, C]
-    img_lr, img_lr_up = [to_tensor_norm(x).float() for x in [img_lr, img_lr_up]]
-
-    img_lr = torch.unsqueeze(img_lr, dim=0)
-    img_lr_up = torch.unsqueeze(img_lr_up, dim=0)
-
-    return img_lr, img_lr_up
-
-
-def load_checkpoint(model, ckpt_path):
-    checkpoint = torch.load(ckpt_path, map_location='cpu')
-    print(f'loding check from: {ckpt_path}')
-    stat_dict = checkpoint['state_dict']['model']
-
-    new_state_dict = OrderedDict()
-    for k, v in stat_dict.items():
-        if k[:7] == 'module.':
-            k = k[7:]  # 去掉 `module.`
-        new_state_dict[k] = v
-
-    model.load_state_dict(new_state_dict)
-    model.cuda()
-    del checkpoint
-    torch.cuda.empty_cache()
-
-
-def model_init(ckpt_path):
-    set_hparams()
-
-    from sam_diffsr.tasks.srdiff_df2k_sam import SRDiffDf2k_sam as trainer
-
-    trainer = trainer()
-
-    trainer.build_model()
-    load_checkpoint(trainer.model, ckpt_path)
-
-    torch.backends.cudnn.benchmark = False
-
-    return trainer
-
-
-def image_infer(img_PIL):
-    with torch.no_grad():
-        trainer.model.eval()
-        img_lr, img_lr_up = get_img_data(img_PIL, hparams, sr_scale=4)
-
-        img_lr = img_lr.to('cuda')
-        img_lr_up = img_lr_up.to('cuda')
-
-        img_sr, _ = trainer.model.sample(img_lr, img_lr_up, img_lr_up.shape)
-
-        img_sr = img_sr.clamp(-1, 1)
-        img_sr = trainer.tensor2img(img_sr)[0]
-        img_sr = Image.fromarray(img_sr)
-
-    return img_sr
-
+from SwinIR.infer import SwinIRDemo
+from sam_diffsr.infer import sam_diffsr_demo
 
-root_path = os.path.dirname(__file__)
 
-cheetah = os.path.join(root_path, "images/0801x4.png")
-print(cheetah)
+def image_infer(img_PIL, progress= gr.Progress(track_tqdm=True)):
+    sam_diffsr_img = sam_diffsr_infer.infer(img_PIL)
+    swin_ir_img = swin_ir_infer.infer(img_PIL)
+    return sam_diffsr_img, swin_ir_img
 
-ckpt_path = os.path.join(root_path, 'sam_diffsr/weight/model_ckpt_steps_400000.ckpt')
-trainer = model_init(ckpt_path)
-demo = gr.Interface(image_infer, gr.Image(type="pil", value=cheetah), "image",
-                    # flagging_options=["blurry", "incorrect", "other"],
-                    examples=[
-                        os.path.join(root_path, "images/0801x4.png"),
-                        os.path.join(root_path, "images/0804x4.png"),
-                        os.path.join(root_path, "images/0809x4.png"),
-                    ]
-                    )
 
 if __name__ == "__main__":
+    sam_diffsr_infer = sam_diffsr_demo()
+    swin_ir_infer = SwinIRDemo()
+    
+    root_path = os.path.dirname(__file__)
+    cheetah = os.path.join(root_path, "images/04011.png")
+    
+    demo = gr.Interface(image_infer, gr.Image(type="pil", value=cheetah),
+                        [
+                                gradio.Image(label='SAM-DiffSR', show_label=True),
+                                gradio.Image(label='SwinIR', show_label=True)
+                        ],
+                        # flagging_options=["blurry", "incorrect", "other"],
+                        examples=[
+                                os.path.join(root_path, "images/04011.png"),
+                                os.path.join(root_path, "images/04033.png"),
+                                os.path.join(root_path, "images/04064.png"),
+                                os.path.join(root_path, "images/04146.png"),
+                                # os.path.join(root_path, "images/10091.png"),
+                                os.path.join(root_path, "images/0801x4.png"),
+                                os.path.join(root_path, "images/0804x4.png"),
+                                os.path.join(root_path, "images/0809x4.png"),
+                        ]
+                        )
+    
     demo.launch()

sam_diffsr/cache/hub/checkpoints/alexnet-owt-7be5be79.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7be5be791159472b1fbf3c69796f7cb30dca7ad8466c2df70058c37116cdee02
+size 244408911

sam_diffsr/infer.py

@@ -0,0 +1,86 @@
+import os
+from collections import OrderedDict
+
+import numpy as np
+import torch
+from PIL import Image
+from torchvision.transforms import transforms
+
+from sam_diffsr.utils_sr.hparams import set_hparams, hparams
+from sam_diffsr.utils_sr.matlab_resize import imresize
+from sam_diffsr.tasks.srdiff_df2k_sam import SRDiffDf2k_sam as trainer_ori
+
+
+ROOT_PATH = os.path.dirname(__file__)
+
+
+class sam_diffsr_demo:
+    def __init__(self):
+        set_hparams()
+        ckpt_path = os.path.join(ROOT_PATH, 'weight/model_ckpt_steps_400000.ckpt')
+        self.model_init(ckpt_path)
+    
+    def get_img_data(self, img_PIL, hparams, sr_scale=4):
+        img_lr = img_PIL.convert('RGB')
+        img_lr = np.uint8(np.asarray(img_lr))
+        
+        h, w, c = img_lr.shape
+        h, w = h * sr_scale, w * sr_scale
+        h = h - h % (sr_scale * 2)
+        w = w - w % (sr_scale * 2)
+        h_l = h // sr_scale
+        w_l = w // sr_scale
+        
+        img_lr = img_lr[:h_l, :w_l]
+        
+        to_tensor_norm = transforms.Compose([
+                transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
+        ])
+        
+        img_lr_up = imresize(img_lr / 256, hparams['sr_scale'])  # np.float [H, W, C]
+        img_lr, img_lr_up = [to_tensor_norm(x).float() for x in [img_lr, img_lr_up]]
+        
+        img_lr = torch.unsqueeze(img_lr, dim=0)
+        img_lr_up = torch.unsqueeze(img_lr_up, dim=0)
+        
+        return img_lr, img_lr_up
+    
+    def load_checkpoint(self, ckpt_path):
+        checkpoint = torch.load(ckpt_path, map_location='cpu')
+        print(f'loding check from: {ckpt_path}')
+        stat_dict = checkpoint['state_dict']['model']
+        
+        new_state_dict = OrderedDict()
+        for k, v in stat_dict.items():
+            if k[:7] == 'module.':
+                k = k[7:]  # 去掉 `module.`
+            new_state_dict[k] = v
+        
+        self.model.model.load_state_dict(new_state_dict)
+        self.model.model.cuda()
+        del checkpoint
+        torch.cuda.empty_cache()
+    
+    def model_init(self, ckpt_path):
+        self.model = trainer_ori()
+        
+        self.model.build_model()
+        self.load_checkpoint(ckpt_path)
+        
+        torch.backends.cudnn.benchmark = False
+    
+    def infer(self, img_PIL):
+        with torch.no_grad():
+            self.model.model.eval()
+            img_lr, img_lr_up = self.get_img_data(img_PIL, hparams, sr_scale=4)
+            
+            img_lr = img_lr.to('cuda')
+            img_lr_up = img_lr_up.to('cuda')
+            
+            img_sr, _ = self.model.model.sample(img_lr, img_lr_up, img_lr_up.shape)
+            
+            img_sr = img_sr.clamp(-1, 1)
+            img_sr = self.model.tensor2img(img_sr)[0]
+            img_sr = Image.fromarray(img_sr)
+        
+        return img_sr

sam_diffsr/models_sr/diffusion.py

@@ -7,7 +7,7 @@ from tqdm import tqdm
 
 from sam_diffsr.utils_sr.plt_img import plt_tensor_img
 from .module_util import default
-from sam_diffsr.utils_sr.sr_utils import SSIM, PerceptualLoss
+from sam_diffsr.utils_sr.sr_utils import SSIM
 from sam_diffsr.utils_sr.hparams import hparams
 
 

sam_diffsr/models_sr/diffusion_sam.py

@@ -52,7 +52,7 @@ class GaussianDiffusion_sam(GaussianDiffusion):
         return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, x0_pred
 
     @torch.no_grad()
-    def sample(self, img_lr, img_lr_up, shape, sam_mask=None, save_intermediate=False):
+    def sample(self, img_lr, img_lr_up, shape, sam_mask=None, save_intermediate=False, progress=None):
         device = self.betas.device
         b = shape[0]