File size: 2,225 Bytes
1675c65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
import torch
from torch.nn import functional as F
from PIL import Image
import numpy as np
import cv2

from rrdbnet_arch import RRDBNet
from utils_sr import *


class RealESRGAN:
    def __init__(self, device, scale=4):
        self.device = device
        self.scale = scale
        self.model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=scale)
        
    def load_weights(self, model_path):
        loadnet = torch.load(model_path)
        if 'params' in loadnet:
            self.model.load_state_dict(loadnet['params'], strict=True)
        elif 'params_ema' in loadnet:
            self.model.load_state_dict(loadnet['params_ema'], strict=True)
        else:
            self.model.load_state_dict(loadnet, strict=True)
        self.model.eval()
        self.model.to(self.device)
        
    @torch.cuda.amp.autocast()
    def predict(self, lr_image, batch_size=4, patches_size=192,
                padding=24, pad_size=15):
        scale = self.scale
        device = self.device
        lr_image = np.array(lr_image)
        lr_image = pad_reflect(lr_image, pad_size)

        patches, p_shape = split_image_into_overlapping_patches(lr_image, patch_size=patches_size, 
                                                                padding_size=padding)
        img = torch.FloatTensor(patches/255).permute((0,3,1,2)).to(device).detach()

        with torch.no_grad():
            res = self.model(img[0:batch_size])
            for i in range(batch_size, img.shape[0], batch_size):
                res = torch.cat((res, self.model(img[i:i+batch_size])), 0)

        sr_image = res.permute((0,2,3,1)).clamp_(0, 1).cpu()
        np_sr_image = sr_image.numpy()

        padded_size_scaled = tuple(np.multiply(p_shape[0:2], scale)) + (3,)
        scaled_image_shape = tuple(np.multiply(lr_image.shape[0:2], scale)) + (3,)
        np_sr_image = stich_together(np_sr_image, padded_image_shape=padded_size_scaled,
                                target_shape=scaled_image_shape, padding_size=padding * scale)
        sr_img = (np_sr_image*255).astype(np.uint8)
        sr_img = unpad_image(sr_img, pad_size*scale)
        sr_img = Image.fromarray(sr_img)

        return sr_img