main.py

from image_resizer import image_preprocess_resizing, image_postprocess_resizing
import cv2
import numpy as np
import onnxruntime as rt

model_path="pretrained_models/RealESRGAN_ANIME_6B_512x512.onnx"

sess = rt.InferenceSession(model_path, providers=['CPUExecutionProvider'])

input_name = sess.get_inputs()[0].name

img = cv2.imread("input.jpg", cv2.IMREAD_UNCHANGED)
img, padding, old_dims = image_preprocess_resizing(img, square_size=512) # 512 is the size of the image you want to resize to (512x512)

h_input, w_input = img.shape[0:2]

img = img.astype(np.float32)

if np.max(img) > 256:  # 16-bit image
    max_range = 65535
    print('Input is a 16-bit image')
else:
    print('Input is not a 16-bit image resetting to 8-bit...')
    max_range = 255
    
# Normalize the image
img = img / max_range
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img = np.transpose(img, (2, 0, 1)).astype(np.float16)

# inference
pred = sess.run(None, {input_name: img[None, ...]})[0]

# Convert back to 8-bit image
outscale = 4
output_img = pred[0].transpose(1,2,0)

# Denormalize the image
if max_range == 65535:  # 16-bit image
    output = (output_img * 65535.0).round().astype(np.uint16)
else:
    output = (output_img * 255.0).round().astype(np.uint8)

output = cv2.resize(
                output, (
                    int(w_input * outscale),
                    int(h_input * outscale),
                ), interpolation=cv2.INTER_LANCZOS4)

# Save the output image
output = cv2.cvtColor(output, cv2.COLOR_RGB2BGR)
output = image_postprocess_resizing(output, padding, old_dims, outscale) # Here you can change the upscale factor default:1

cv2.imwrite("result.jpg", output)