Nicole-Yi's picture
Upload 103 files
4a5e815 verified
import argparse
import numpy as np
from PIL import Image
import json
import os
from datetime import datetime
from scipy.ndimage import sobel
from skimage.metrics import structural_similarity as ssim
from skimage.metrics import peak_signal_noise_ratio as psnr
def compute_edge_intensity(img_array):
"""
Calculate edge intensity (Sobel gradient) of an image.
Args:
img_array (np.ndarray): Input image (RGB or grayscale)
Returns:
np.ndarray: Edge intensity map
"""
if len(img_array.shape) == 3:
gray = np.mean(img_array, axis=2).astype(np.float32)
else:
gray = img_array.astype(np.float32)
grad_x = sobel(gray, axis=1)
grad_y = sobel(gray, axis=0)
edge_intensity = np.sqrt(grad_x ** 2 + grad_y ** 2)
return edge_intensity
def evaluate_super_resolution(input_path, output_path):
"""
Evaluate super-resolution quality for RRDN GANS model, supporting 2x or 4x scaling.
Args:
input_path (str): Path to input low-resolution image (e.g., baboon.png)
output_path (str): Path to output super-resolution image (e.g., output_baboon_gans.jpg)
Returns:
dict: Evaluation results containing PSNR, ESI, SSIM and success status
"""
# Load images
input_img = Image.open(input_path).convert('RGB')
output_img = Image.open(output_path).convert('RGB')
# Convert to numpy arrays
input_array = np.array(input_img)
output_array = np.array(output_img)
# Check resolution (accept 2x or 4x scaling)
input_h, input_w = input_array.shape[:2]
output_h, output_w = output_array.shape[:2]
valid_resolution = (output_h == 2 * input_h and output_w == 2 * input_w) or \
(output_h == 4 * input_h and output_w == 4 * input_w)
if not valid_resolution:
return {
'PSNR': 0.0,
'PSNR Threshold': 20.0,
'ESI (Edge Strength Improvement)': 0.0,
'ESI Threshold': 1.1,
'SSIM': 0.0,
'SSIM Threshold': 0.5,
'Success': False,
'Error': f"Output resolution ({output_w}x{output_h}) is neither 2x nor 4x input ({input_w}x{input_h})"
}
# Determine scaling factor
scale = 4 if output_h == 4 * input_h else 2
# Upsample input image to output resolution (bilinear interpolation) as pseudo-reference
upsampled_img = input_img.resize((output_w, output_h), Image.BILINEAR)
upsampled_array = np.array(upsampled_img)
# Calculate PSNR
psnr_value = psnr(upsampled_array, output_array, data_range=255)
# Calculate edge intensity
input_edge = compute_edge_intensity(upsampled_array)
output_edge = compute_edge_intensity(output_array)
# Save edge intensity maps for debugging
input_edge_img = Image.fromarray((input_edge / input_edge.max() * 255).astype(np.uint8))
output_edge_img = Image.fromarray((output_edge / output_edge.max() * 255).astype(np.uint8))
input_edge_img.save('input_edge.png')
output_edge_img.save('output_edge.png')
# Calculate ESI (Edge Strength Improvement)
input_edge_mean = np.mean(input_edge)
output_edge_mean = np.mean(output_edge)
esi = output_edge_mean / (input_edge_mean + 1e-10)
# Calculate SSIM
ssim_value = ssim(upsampled_array, output_array, channel_axis=2, data_range=255)
# Thresholds adjusted dynamically based on scaling factor
psnr_threshold = 20.0 if scale == 2 else 18.0
esi_threshold = 1.1
ssim_threshold = 0.5
success = psnr_value >= psnr_threshold and esi >= esi_threshold and ssim_value >= ssim_threshold
return {
'Scale': scale,
'PSNR': psnr_value,
'PSNR Threshold': psnr_threshold,
'ESI (Edge Strength Improvement)': esi,
'ESI Threshold': esi_threshold,
'SSIM': ssim_value,
'SSIM Threshold': ssim_threshold,
'Success': success
}
def validate_inputs(input_path, output_path):
"""
Validate input files exist, are non-empty, and have correct format.
"""
try:
if not os.path.exists(input_path):
return False, f"Input file {input_path} does not exist."
if not os.path.exists(output_path):
return False, f"Output file {output_path} does not exist."
if os.path.getsize(input_path) == 0:
return False, f"Input file {input_path} is empty."
if os.path.getsize(output_path) == 0:
return False, f"Output file {output_path} is empty."
try:
with Image.open(input_path) as img:
img.verify()
with Image.open(output_path) as img:
img.verify()
except Exception as e:
return False, f"Invalid image format: {str(e)}"
return True, ""
except Exception as e:
return False, f"Validation error: {str(e)}"
def save_result_to_jsonl(result_path, process, result, comments):
"""
Save results to JSONL file.
"""
time_point = datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
record = {
"Process": process,
"Result": bool(result),
"TimePoint": time_point,
"comments": comments
}
try:
with open(result_path, 'a', encoding='utf-8') as f:
json_line = json.dumps(record, ensure_ascii=False)
f.write(json_line + '\n')
except Exception as e:
print(f"Failed to save result to {result_path}: {str(e)}")
def main():
parser = argparse.ArgumentParser(
description='Evaluate super-resolution quality for RRDN GANS model, supporting 2x or 4x scaling.')
parser.add_argument('--groundtruth', type=str, required=True,
help='Path to input low-resolution image (e.g., baboon.png)')
parser.add_argument('--output', type=str, required=True,
help='Path to output super-resolution image (e.g., output_baboon_gans.jpg)')
parser.add_argument('--result', type=str, help='Path to JSONL file to save results')
args = parser.parse_args()
process, comments = validate_inputs(args.groundtruth, args.output)
result_dict = {}
success = False
if process:
try:
result_dict = evaluate_super_resolution(args.groundtruth, args.output)
success = result_dict.get('Success', False)
if 'Error' in result_dict:
comments = result_dict['Error']
success = False
else:
comments = (
f"Scale: {result_dict['Scale']}x, "
f"PSNR: {result_dict['PSNR']:.3f} "
f"(Threshold: {result_dict['PSNR Threshold']}), "
f"ESI: {result_dict['ESI (Edge Strength Improvement)']:.3f} "
f"(Threshold: {result_dict['ESI Threshold']}), "
f"SSIM: {result_dict['SSIM']:.3f} "
f"(Threshold: {result_dict['SSIM Threshold']}), "
f"Success: {success}"
)
if not success:
comments += ". Possible issues: Low PSNR (poor pixel similarity), low ESI (insufficient edge enhancement), or low SSIM (structural distortion)."
except Exception as e:
process = False
success = False
comments = f"Evaluation failed: {str(e)}"
else:
success = False
print('Evaluation Results:')
if process and 'Error' not in result_dict:
for key, value in result_dict.items():
if key != 'Error':
print(f'{key}: {value}')
else:
print(f'Error: {comments}')
if args.result:
save_result_to_jsonl(args.result, process, success, comments)
if __name__ == '__main__':
main()