Extracting Menu Images

Bash
# 1. Fetch the Uber Eats page with browser-like headers
curl -s -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15" \
  "https://www.ubereats.com/fr-en/store/restaurant-name/store-id" \
  -o /tmp/ubereats.html

# 2. Run the extraction script (fetches, extracts URLs, downloads images, pulls JSON-LD menu)
python3 scripts/extract-ubereats.py \
  "https://www.ubereats.com/fr-en/store/restaurant-name/store-id" \
  /tmp/output_folder

# 3. Identify images (use parallel subagents for 50+ images) then build the mapping

Progress:
- [ ] Fetch page HTML with browser-like headers
- [ ] Extract unique image URLs (regex on CDN pattern)
- [ ] Extract JSON-LD menu data (names, sections, prices)
- [ ] Download all images to a local folder
- [ ] Identify each image (direct Read, or parallel Task subagents if 50+)
- [ ] Build image-to-item mapping (products AND options)
- [ ] Update catalog JSON with `imageUrl` fields
- [ ] Spot-check a few images against the live site

1. Fetch the page Use a real browser User-Agent and Accept-Language headers to avoid rate limiting. Uber Eats pages are React-rendered; the HTML still contains embedded image URLs and JSON-LD data even though live correlation requires JS.

2. Extract image URLs

Python
import re
pattern = r'tb-static\.uber\.com/prod/image-proc/processed_images/([a-f0-9]+)/([a-f0-9]+)\.(?:jpeg|png)'
matches = re.findall(pattern, html)
seen, images = set(), []
for item_hash, size_hash in matches:
    if item_hash not in seen:
        seen.add(item_hash)
        images.append(f'https://tb-static.uber.com/prod/image-proc/processed_images/{item_hash}/{size_hash}.jpeg')

For non-Uber-Eats sites, check src, data-src, srcset, and alt attributes, and common CDN domains (Webflow cdn.prod.website-files.com, Squarespace, Cloudinary).

3. Extract JSON-LD menu data (Uber Eats only) — parses <script type="application/ld+json"> blocks for Restaurant → hasMenu → hasMenuSection → hasMenuItem, giving accurate names/prices to correlate against.

4. Download images to a numbered local folder (01.jpeg, 02.jpeg, ...) with a 0.3–0.5s delay between requests.

5. Identify images

• Small sets (<50): Read each image directly and note what it shows.
• Large sets (50+): launch parallel Task subagents, each covering a 20–30 image batch, returning a JSON mapping of image number → product name. Combine results after all agents finish.

6. Build the mapping and update the catalog Assign imageUrl to both menu products and options (drinks, sauces, sides, desserts) — option images are often present in the extraction but get overlooked.