File size: 8,041 Bytes
bac741f | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 | """
SSIM + Bootstrap CI computation for all benchmark methods.
Works with existing rendered screenshots and per-sample CLIP scores.
Usage:
python scripts/step_ssim_bootstrap.py --benchmark_dir results/benchmark --ref_dir data/ref_screenshots
python scripts/step_ssim_bootstrap.py --benchmark_dir results/benchmark_websight --ref_dir data/ref_screenshots_websight
"""
import argparse
import json
import os
import sys
from pathlib import Path
import numpy as np
from PIL import Image
PROJECT_ROOT = Path(__file__).parent.parent
def compute_ssim_pil(img1, img2, win_size=7):
"""Compute SSIM between two PIL images using numpy (no skimage dependency)."""
target_size = (min(img1.width, img2.width, 512), min(img1.height, img2.height, 512))
a = np.array(img1.resize(target_size).convert("RGB"), dtype=np.float64)
b = np.array(img2.resize(target_size).convert("RGB"), dtype=np.float64)
C1 = (0.01 * 255) ** 2
C2 = (0.03 * 255) ** 2
ssims = []
for ch in range(3):
mu1 = uniform_filter(a[:, :, ch], win_size)
mu2 = uniform_filter(b[:, :, ch], win_size)
mu1_sq = mu1 ** 2
mu2_sq = mu2 ** 2
mu1_mu2 = mu1 * mu2
sigma1_sq = uniform_filter(a[:, :, ch] ** 2, win_size) - mu1_sq
sigma2_sq = uniform_filter(b[:, :, ch] ** 2, win_size) - mu2_sq
sigma12 = uniform_filter(a[:, :, ch] * b[:, :, ch], win_size) - mu1_mu2
ssim_map = ((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / \
((mu1_sq + mu2_sq + C1) * (sigma1_sq + sigma2_sq + C2))
ssims.append(ssim_map.mean())
return float(np.mean(ssims))
def uniform_filter(arr, size):
"""Simple uniform (box) filter."""
from scipy.ndimage import uniform_filter as _uf
return _uf(arr, size=size, mode='reflect')
def render_html_to_screenshot(html_path, out_path, width=1280, height=1024, timeout=15000):
"""Render HTML file to PNG screenshot using Playwright."""
try:
from playwright.sync_api import sync_playwright
with sync_playwright() as p:
browser = p.chromium.launch(headless=True, args=['--no-sandbox', '--disable-gpu'])
page = browser.new_page(viewport={"width": width, "height": height})
page.goto(f"file://{html_path}", wait_until="networkidle", timeout=timeout)
page.wait_for_timeout(1000)
page.screenshot(path=str(out_path), full_page=True)
browser.close()
return True
except Exception as e:
print(f" Render failed for {html_path}: {e}")
return False
def compute_ssim_for_method(method_dir, ref_dir, render_cache_dir):
"""Compute per-sample SSIM for a method."""
html_dir = Path(method_dir) / "html_predictions"
if not html_dir.exists():
return None
ref_dir = Path(ref_dir)
render_dir = Path(render_cache_dir) / Path(method_dir).name
render_dir.mkdir(parents=True, exist_ok=True)
html_files = sorted(html_dir.glob("*.html"))
per_sample = {}
for html_path in html_files:
sample_id = html_path.stem
ref_path = ref_dir / f"{sample_id}.png"
if not ref_path.exists():
continue
rendered_path = render_dir / f"{sample_id}.png"
if not rendered_path.exists():
ok = render_html_to_screenshot(str(html_path.resolve()), str(rendered_path))
if not ok:
continue
try:
ref_img = Image.open(ref_path).convert("RGB")
rendered_img = Image.open(rendered_path).convert("RGB")
ssim = compute_ssim_pil(ref_img, rendered_img)
per_sample[sample_id] = ssim
except Exception as e:
print(f" SSIM error for {sample_id}: {e}")
continue
if not per_sample:
return None
return {
"n_samples": len(per_sample),
"avg_ssim": round(float(np.mean(list(per_sample.values()))), 4),
"std_ssim": round(float(np.std(list(per_sample.values()))), 4),
"per_sample": per_sample,
}
def bootstrap_ci(scores, n_bootstrap=10000, ci=0.95, seed=42):
"""Compute bootstrap confidence interval."""
rng = np.random.RandomState(seed)
scores = np.array(scores)
n = len(scores)
boot_means = np.array([
rng.choice(scores, size=n, replace=True).mean()
for _ in range(n_bootstrap)
])
alpha = (1 - ci) / 2
lo = float(np.percentile(boot_means, 100 * alpha))
hi = float(np.percentile(boot_means, 100 * (1 - alpha)))
return {
"mean": float(scores.mean()),
"ci_lower": round(lo, 4),
"ci_upper": round(hi, 4),
"ci_width": round(hi - lo, 4),
"std": round(float(scores.std()), 4),
"n": n,
}
def compute_bootstrap_for_all(benchmark_dir):
"""Compute bootstrap CI for all methods from per-sample CLIP scores."""
benchmark_dir = Path(benchmark_dir)
results = {}
for method_dir in sorted(benchmark_dir.iterdir()):
if not method_dir.is_dir():
continue
clip_file = method_dir / "clip_scores.json"
if not clip_file.exists():
continue
with open(clip_file) as f:
clip_data = json.load(f)
per_sample = clip_data.get("per_sample", {})
if not per_sample:
continue
scores = []
for k, v in per_sample.items():
if isinstance(v, dict):
scores.append(v.get("clip_score", 0))
else:
scores.append(float(v))
if not scores:
continue
ci_result = bootstrap_ci(scores)
results[method_dir.name] = ci_result
print(f" {method_dir.name}: CLIP={ci_result['mean']:.4f} [{ci_result['ci_lower']:.4f}, {ci_result['ci_upper']:.4f}]")
return results
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--benchmark_dir", type=str, default=str(PROJECT_ROOT / "results" / "benchmark"))
parser.add_argument("--ref_dir", type=str, default=str(PROJECT_ROOT / "data" / "ref_screenshots"))
parser.add_argument("--render_cache", type=str, default=str(PROJECT_ROOT / "results" / "rendered_screenshots"))
parser.add_argument("--skip_ssim", action="store_true")
parser.add_argument("--skip_bootstrap", action="store_true")
args = parser.parse_args()
benchmark_dir = Path(args.benchmark_dir)
output = {}
if not args.skip_bootstrap:
print("=" * 60)
print("Computing Bootstrap CI for CLIP scores...")
print("=" * 60)
bootstrap_results = compute_bootstrap_for_all(args.benchmark_dir)
output["bootstrap_ci"] = bootstrap_results
ci_file = benchmark_dir / "bootstrap_ci.json"
with open(ci_file, "w") as f:
json.dump(bootstrap_results, f, indent=2)
print(f"\nSaved to {ci_file}")
if not args.skip_ssim:
print("\n" + "=" * 60)
print("Computing SSIM scores...")
print("=" * 60)
ssim_results = {}
for method_dir in sorted(benchmark_dir.iterdir()):
if not method_dir.is_dir():
continue
html_dir = method_dir / "html_predictions"
if not html_dir.exists():
continue
print(f"\n Processing {method_dir.name}...")
result = compute_ssim_for_method(str(method_dir), args.ref_dir, args.render_cache)
if result:
ssim_results[method_dir.name] = {
"n_samples": result["n_samples"],
"avg_ssim": result["avg_ssim"],
"std_ssim": result["std_ssim"],
}
print(f" SSIM={result['avg_ssim']:.4f} ± {result['std_ssim']:.4f} (n={result['n_samples']})")
output[f"ssim_{method_dir.name}"] = result
ssim_file = benchmark_dir / "ssim_scores.json"
with open(ssim_file, "w") as f:
json.dump(ssim_results, f, indent=2)
print(f"\nSaved to {ssim_file}")
if __name__ == "__main__":
main()
|