Moose Finder

Automatically finds photos of Moose (a Bernese Mountain Dog) from Pet Camp's daily camper cameos.

How It Works

1. Scrapes the daily "Main Campground & Ranger Station" photos from campercameos.com
2. Detects individual dogs in each photo using YOLOv8n
3. Compares each dog crop (plus the full image) against reference photos of Moose using MegaDescriptor embeddings
4. Classifies results as match (>90%), uncertain (50-90%), or no match (<50%)

MegaDescriptor is a foundation model specifically designed for individual animal re-identification, trained on dog-specific datasets. Combined with YOLOv8n crop detection, it reliably identifies Moose even in multi-dog photos.

Setup

1. Install Dependencies

pip install -r requirements.txt

2. Add Reference Photos

Place clear photos of Moose in reference_photos/. More photos covering different poses and angles improve accuracy. The current set includes 16 reference photos covering front-face, full body, side view, and outdoor poses.

3. Run

python moose_finder.py

Usage

# Check today's photos
python moose_finder.py

# Check a specific date
python moose_finder.py --date 2026-02-01

# Download images without analyzing
python moose_finder.py --download-only

# Only analyze already-cached images (no downloads)
python moose_finder.py --analyze-only

# Limit new downloads
python moose_finder.py --download-limit 10

# Show local status for a date (no downloads or analysis)
python moose_finder.py --show-results
python moose_finder.py --show-results --date 2026-02-03

# Debug a single photo with full diagnostic breakdown
python moose_finder.py --url "https://petcamp.s3.us-west-1.amazonaws.com/..."

Debug Mode

The --url flag analyzes a single photo and shows the full diagnostic breakdown: per-reference similarity scores, YOLO detections with bounding boxes, crop dimensions, and saves all dog crops to debug/ for visual inspection.

Full image similarity: 21% (best ref: moose_ref_4)
  moose_ref_1:     6%
  moose_ref_2:    21%
  ...

YOLOv8n detected 4 dog(s):

  Dog 1: bbox [703, 593, 1177, 1105] conf=0.94, crop 559x614
    Similarity: 26% (best ref: moose_ref_2)

  Dog 3: bbox [118, 552, 608, 988] conf=0.83, crop 588x522
    Similarity: 94% (best ref: moose_ref_15)  <- best

Best overall: 94% (from detection_3) -> match

Tips

• Pet Camp usually posts photos in the afternoon/evening (SF time)
• Add more reference photos for better accuracy, especially for poses where Moose is missed
• Results are saved per-date, so re-running a date skips already-analyzed photos
• Downloaded images are cached in cache/ to avoid re-downloading

Troubleshooting

"No photos found for this date"

• Photos might not be posted yet (usually by evening)
• The URL format may have changed — check campercameos.com manually

"No reference photos found"

• Add photos of Moose to reference_photos/

Moose not detected in a photo

• Use --url to debug the specific photo
• If Moose is small in the background, YOLO may not detect it (known limitation)
• If the crop is good but similarity is low, add the photo as a new reference to cover that pose