File size: 4,248 Bytes
14ce5a9 |
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 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 |
import os
import argparse
from PIL import Image
from tqdm import tqdm
from torchvision import transforms
from torch.utils.data import Dataset, DataLoader
import torch
import torch.nn.functional as F
from ocr import OCR
from character_error_rate import CharacterErrorRate
from word_error_rate import WordErrorRate
from torchmetrics.image import (
PeakSignalNoiseRatio,
StructuralSimilarityIndexMeasure,
LearnedPerceptualImagePatchSimilarity,
FrechetInceptionDistance,
)
class ImageFolderPairDataset(Dataset):
def __init__(self, dir1, dir2, transform=None):
self.dir1 = dir1
self.dir2 = dir2
self.filenames = sorted(os.listdir(dir1))
self.transform = transform
def __len__(self):
return len(self.filenames)
def __getitem__(self, idx):
name = self.filenames[idx]
img1 = Image.open(os.path.join(self.dir1, name)).convert("RGB")
img2 = Image.open(os.path.join(self.dir2, name)).convert("RGB")
if self.transform:
img1 = self.transform(img1)
img2 = self.transform(img2)
return img1, img2
def evaluate(args):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")
transform = transforms.Compose(
[transforms.Resize((args.image_size, args.image_size)), transforms.ToTensor()]
)
dataset = ImageFolderPairDataset(
args.original_dir, args.reconstructed_dir, transform
)
loader = DataLoader(
dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.num_workers
)
if "cer" in args.metrics or "wer" in args.metrics:
ocr = OCR(device)
# Metrics init
metrics = {}
if "psnr" in args.metrics:
metrics["psnr"] = PeakSignalNoiseRatio().to(device)
if "ssim" in args.metrics:
metrics["ssim"] = StructuralSimilarityIndexMeasure().to(device)
if "lpips" in args.metrics:
metrics["lpips"] = LearnedPerceptualImagePatchSimilarity().to(device)
if "fid" in args.metrics:
metrics["fid"] = FrechetInceptionDistance().to(device)
if "cer" in args.metrics:
metrics["cer"] = CharacterErrorRate(ocr)
if "wer" in args.metrics:
metrics["wer"] = WordErrorRate(ocr)
for batch in tqdm(loader, desc="Evaluating"):
# img1, img1_path, img2, img2_path = [b.to(device) for b in batch]
img1, img2 = [b.to(device) for b in batch]
if "psnr" in metrics:
metrics["psnr"].update(img2, img1)
if "ssim" in metrics:
metrics["ssim"].update(img2, img1)
if "lpips" in metrics:
metrics["lpips"].update(img2, img1)
if "cer" in metrics:
metrics["cer"].update(img2, img1)
if "wer" in metrics:
metrics["wer"].update(img2, img1)
if "fid" in metrics:
img1_uint8 = (img1 * 255).clamp(0, 255).to(torch.uint8)
img2_uint8 = (img2 * 255).clamp(0, 255).to(torch.uint8)
metrics["fid"].update(img1_uint8, real=True)
metrics["fid"].update(img2_uint8, real=False)
print("\nResults:")
for name, metric in metrics.items():
print(f"{name.upper()}", end="\t")
print()
for name, metric in metrics.items():
result = metric.compute().item()
print(f"{result:.4f}", end="\t")
print()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--original_dir", type=str, required=True, help="Path to original images"
)
parser.add_argument(
"--reconstructed_dir",
type=str,
required=True,
help="Path to reconstructed images",
)
parser.add_argument(
"--metrics",
nargs="+",
default=["psnr", "ssim", "lpips", "fid"],
help="Metrics to compute: psnr, ssim, lpips, fid",
)
parser.add_argument(
"--batch_size", type=int, default=8, help="Batch size for processing"
)
parser.add_argument("--image_size", type=int, default=256, help="Image resize size")
parser.add_argument(
"--num_workers", type=int, default=4, help="Number of workers for DataLoader"
)
args = parser.parse_args()
evaluate(args)
|