File size: 12,577 Bytes
bb01096 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 | """
UIPress Step 1b: Compute evaluation metrics on generated HTML.
Uses Design2Code's official metrics + our custom token analysis.
Usage:
# Evaluate a single model's predictions
python scripts/step1_eval_metrics.py --pred_dir results/qwen2_5_vl_7b/html_predictions
# Evaluate all models in results/
python scripts/step1_eval_metrics.py --all
# Quick CLIP-only evaluation (much faster, no browser needed)
python scripts/step1_eval_metrics.py --all --clip_only
"""
# ---- HuggingFace 镜像 (必须在其他 import 之前) ----
import os
os.environ["HF_ENDPOINT"] = os.environ.get("HF_ENDPOINT", "https://hf-mirror.com")
os.environ["HF_HOME"] = os.environ.get("HF_HOME", "/root/rivermind-data/huggingface")
import argparse
import json
import sys
import subprocess
import tempfile
from pathlib import Path
import torch
from PIL import Image
from tqdm import tqdm
PROJECT_ROOT = Path(__file__).parent.parent
# ============================================================
# CLIP Score (standalone, no Design2Code repo needed)
# ============================================================
class CLIPScorer:
"""Compute CLIP similarity between reference screenshot and generated HTML rendering."""
def __init__(self, device="cuda"):
import open_clip
self.device = device
self.model, _, self.preprocess = open_clip.create_model_and_transforms(
"ViT-B-32", pretrained="openai" # ViT-B/32 matches Design2Code's metric
)
self.model = self.model.to(device).eval()
@torch.no_grad()
def score_images(self, img1: Image.Image, img2: Image.Image) -> float:
"""Compute CLIP cosine similarity between two images."""
t1 = self.preprocess(img1).unsqueeze(0).to(self.device)
t2 = self.preprocess(img2).unsqueeze(0).to(self.device)
f1 = self.model.encode_image(t1)
f2 = self.model.encode_image(t2)
f1 = f1 / f1.norm(dim=-1, keepdim=True)
f2 = f2 / f2.norm(dim=-1, keepdim=True)
return float((f1 * f2).sum())
# ============================================================
# HTML to Screenshot Rendering
# ============================================================
def render_html_to_image(html_path: str, output_path: str, width: int = 1280, height: int = 1024) -> bool:
"""Render HTML file to screenshot. Tries Playwright first, falls back to Selenium."""
# Try Playwright (Design2Code official uses this)
if _render_playwright(html_path, output_path, width, height):
return True
# Fallback to Selenium
return _render_selenium(html_path, output_path, width, height)
def _render_playwright(html_path: str, output_path: str, width: int, height: int) -> bool:
"""Render using Playwright (recommended, matches Design2Code pipeline)."""
try:
from playwright.sync_api import sync_playwright
abs_path = os.path.abspath(html_path)
with sync_playwright() as p:
browser = p.chromium.launch(headless=True)
page = browser.new_page(viewport={"width": width, "height": height})
page.goto(f"file://{abs_path}", wait_until="networkidle")
page.screenshot(path=output_path, full_page=False)
browser.close()
return True
except ImportError:
return False
except Exception as e:
print(f" [WARN] Playwright render failed for {html_path}: {e}")
return False
def _render_selenium(html_path: str, output_path: str, width: int, height: int) -> bool:
"""Fallback: render using Selenium + Chrome."""
try:
from selenium import webdriver
from selenium.webdriver.chrome.options import Options
options = Options()
options.add_argument("--headless=new")
options.add_argument("--no-sandbox")
options.add_argument("--disable-dev-shm-usage")
options.add_argument(f"--window-size={width},{height}")
options.add_argument("--disable-gpu")
options.add_argument("--force-device-scale-factor=1")
driver = webdriver.Chrome(options=options)
driver.set_window_size(width, height)
abs_path = os.path.abspath(html_path)
driver.get(f"file://{abs_path}")
import time
time.sleep(1)
driver.save_screenshot(output_path)
driver.quit()
return True
except Exception as e:
print(f" [WARN] Selenium render failed for {html_path}: {e}")
return False
# ============================================================
# Evaluation Pipeline
# ============================================================
def evaluate_predictions(pred_dir: str, ref_dir: str, clip_only: bool = False):
"""
Evaluate HTML predictions against reference screenshots.
Args:
pred_dir: Directory containing predicted .html files
ref_dir: Directory containing reference .png screenshots (and .html ground truth)
clip_only: If True, only compute CLIP score (faster)
"""
pred_path = Path(pred_dir)
ref_path = Path(ref_dir)
pred_files = sorted(pred_path.glob("*.html"))
if not pred_files:
print(f"No HTML files found in {pred_dir}")
return None
print(f"Found {len(pred_files)} predictions in {pred_dir}")
# Initialize CLIP scorer
device = "cuda" if torch.cuda.is_available() else "cpu"
clip_scorer = CLIPScorer(device=device)
results = {}
clip_scores = []
with tempfile.TemporaryDirectory() as tmp_dir:
for pred_file in tqdm(pred_files, desc="Evaluating"):
sample_id = pred_file.stem
# Find reference screenshot
ref_img_path = ref_path / f"{sample_id}.png"
if not ref_img_path.exists():
# Try numeric matching
for ext in [".png", ".jpg", ".jpeg"]:
candidate = ref_path / f"{sample_id}{ext}"
if candidate.exists():
ref_img_path = candidate
break
if not ref_img_path.exists():
continue
ref_img = Image.open(ref_img_path).convert("RGB")
# Render predicted HTML to image
pred_img_path = os.path.join(tmp_dir, f"{sample_id}.png")
success = render_html_to_image(str(pred_file), pred_img_path)
if success and os.path.exists(pred_img_path):
pred_img = Image.open(pred_img_path).convert("RGB")
clip_score = clip_scorer.score_images(ref_img, pred_img)
else:
clip_score = 0.0
clip_scores.append(clip_score)
results[sample_id] = {"clip_score": round(clip_score, 4)}
# Summary
avg_clip = sum(clip_scores) / len(clip_scores) if clip_scores else 0
summary = {
"n_evaluated": len(clip_scores),
"avg_clip_score": round(avg_clip, 4),
"min_clip_score": round(min(clip_scores), 4) if clip_scores else 0,
"max_clip_score": round(max(clip_scores), 4) if clip_scores else 0,
"per_sample": results,
}
print(f"\n{'='*50}")
print(f"CLIP Score: {avg_clip:.4f} (n={len(clip_scores)})")
print(f" Min: {summary['min_clip_score']:.4f}")
print(f" Max: {summary['max_clip_score']:.4f}")
print(f"{'='*50}")
return summary
def run_design2code_official_eval(pred_dir: str):
"""
Run Design2Code's official evaluation script if available.
This provides Block-Match, Text, Position, Color metrics.
"""
eval_script = PROJECT_ROOT / "repos" / "Design2Code" / "metrics" / "multi_processing_eval.py"
if not eval_script.exists():
print("[INFO] Design2Code official eval not available.")
print(" Clone the repo first: git clone https://github.com/NoviScl/Design2Code repos/Design2Code")
print(" Only CLIP score was computed.")
return None
print(f"\nRunning Design2Code official evaluation...")
print(f" Script: {eval_script}")
print(f" Predictions: {pred_dir}")
# Note: The official script requires manual configuration of prediction directories
# We print instructions for the user
print(f"\n To run manually:")
print(f" 1. Edit {eval_script} line ~54")
print(f" 2. Set prediction directory to: {pred_dir}")
print(f" 3. Run: python {eval_script}")
return None
# ============================================================
# CLI
# ============================================================
def main():
parser = argparse.ArgumentParser(description="UIPress: Evaluate generated HTML")
parser.add_argument("--pred_dir", type=str, help="Directory with predicted .html files")
parser.add_argument("--ref_dir", type=str, default=None,
help="Directory with reference .png screenshots")
parser.add_argument("--all", action="store_true",
help="Evaluate all models in results/")
parser.add_argument("--clip_only", action="store_true",
help="Only compute CLIP score (skip Design2Code metrics)")
parser.add_argument("--output", type=str, default=None,
help="Output JSON path")
args = parser.parse_args()
# Find reference directory
if args.ref_dir is None:
candidates = [
PROJECT_ROOT / "data" / "testset_final",
PROJECT_ROOT / "repos" / "Design2Code" / "testset_final",
]
for c in candidates:
if c.exists():
args.ref_dir = str(c)
break
# Fallback: extract reference images from HF dataset
if args.ref_dir is None:
hf_path = PROJECT_ROOT / "data" / "design2code"
if hf_path.exists():
ref_tmp = PROJECT_ROOT / "data" / "ref_screenshots"
ref_tmp.mkdir(parents=True, exist_ok=True)
if not any(ref_tmp.iterdir()):
print("Extracting reference screenshots from HF dataset...")
from datasets import load_from_disk
ds = load_from_disk(str(hf_path))
if hasattr(ds, 'keys'):
ds = ds[list(ds.keys())[0]]
for i, item in enumerate(ds):
img = item.get("image") or item.get("screenshot")
if img is not None:
if not isinstance(img, Image.Image):
img = Image.open(img).convert("RGB")
img.save(str(ref_tmp / f"{i}.png"))
print(f" Extracted {len(list(ref_tmp.glob('*.png')))} screenshots")
args.ref_dir = str(ref_tmp)
if args.ref_dir is None:
print("Cannot find reference data. Specify --ref_dir or run download_data.py")
sys.exit(1)
if args.all:
# Evaluate all models
results_dir = PROJECT_ROOT / "results"
all_results = {}
for model_dir in sorted(results_dir.iterdir()):
pred_html = model_dir / "html_predictions"
if pred_html.exists():
print(f"\n{'='*60}")
print(f"Evaluating: {model_dir.name}")
print(f"{'='*60}")
summary = evaluate_predictions(
str(pred_html), args.ref_dir, args.clip_only
)
if summary:
all_results[model_dir.name] = summary
# Print comparison
if all_results:
print(f"\n{'='*70}")
print("FINAL COMPARISON")
print(f"{'='*70}")
print(f"{'Model':<30} {'CLIP Score':>12} {'N Samples':>12}")
print("-" * 55)
for name, s in sorted(all_results.items(), key=lambda x: -x[1]["avg_clip_score"]):
print(f"{name:<30} {s['avg_clip_score']:>12.4f} {s['n_evaluated']:>12}")
# Save
out = results_dir / "step1_metrics_comparison.json"
with open(out, "w") as f:
json.dump(all_results, f, indent=2)
print(f"\nSaved to: {out}")
elif args.pred_dir:
summary = evaluate_predictions(args.pred_dir, args.ref_dir, args.clip_only)
if summary and args.output:
with open(args.output, "w") as f:
json.dump(summary, f, indent=2)
# Also try official eval
if not args.clip_only:
run_design2code_official_eval(args.pred_dir)
else:
parser.print_help()
if __name__ == "__main__":
main()
|