🛒 Retail Shelf Stock Monitor
CLIP gates empty shelves → Grounding DINO detects products → post-processing gives stock level + misplaced flag
1. CLIP (zero-shot classification) – decides if the shelf is empty.
2. Grounding DINO – detects grocery items with bounding boxes.
3. Post-processing – filters false positives, deduplicates, counts products, and flags misplaced items via spatial outlier detection.
4. Output – annotated image/video with stock status and misplaced highlights.
CLIP acts as a fast gate → only non-empty shelves trigger object detection.
Record a short webcam clip. The app analyzes one completed clip at a time and waits before the next run.
📌 About This Demo
Combines CLIP and Grounding DINO to estimate shelf stock, count products, and flag possible misplaced items.
⚡ Runtime Note
Runs on CPU → slower for videos. For a faster GPU-backed walkthrough, use the Colab notebook.
👤 Author
Suresh Lama
Website · Hugging Face · GitHub