Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: educational-use-only
|
| 4 |
+
license_link: https://github.com/hongming111/SNAIC_Week4_SmartCart/blob/main/LICENSE
|
| 5 |
+
tags:
|
| 6 |
+
- computer-vision
|
| 7 |
+
- object-detection
|
| 8 |
+
- image-classification
|
| 9 |
+
- gradio
|
| 10 |
+
- yolo
|
| 11 |
+
- dinov2
|
| 12 |
+
- onnx
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# SmartCart SG — trained bundle (SNAIC Week 4 Capstone)
|
| 16 |
+
|
| 17 |
+
Trained artifacts for a self-checkout computer-vision pipeline (detect -> recognize -> price -> receipt)
|
| 18 |
+
built on a home-grown Singapore grocery dataset. Source code and full writeup:
|
| 19 |
+
[hongming111/SNAIC_Week4_SmartCart](https://github.com/hongming111/SNAIC_Week4_SmartCart).
|
| 20 |
+
|
| 21 |
+
Powers the companion Space: [Hongming111/SNAIC_Week4_SmartCart](https://huggingface.co/spaces/Hongming111/SNAIC_Week4_SmartCart).
|
| 22 |
+
|
| 23 |
+
## Contents
|
| 24 |
+
|
| 25 |
+
| File | Purpose |
|
| 26 |
+
|---|---|
|
| 27 |
+
| `detector.pt`, `detector_v0.pt` | YOLO11 product detector (basket-hardened / earlier version) |
|
| 28 |
+
| `FastSAM-s.pt` | FastSAM "segment everything" proposals for Basket mode |
|
| 29 |
+
| `head.pt`, `head.onnx`, `head.int8.onnx` | 25-way recognition head on frozen DINOv2 features (PyTorch / ONNX / quantized) |
|
| 30 |
+
| `gallery_index.npy`, `gallery_meta.csv` | DINOv2 embedding gallery used for open-set gating |
|
| 31 |
+
| `catalog_prices.csv` | Real median FairPrice SGD prices per class |
|
| 32 |
+
| `confidence_threshold.json` | Calibrated serving threshold (0.6, held-out crop split) |
|
| 33 |
+
| `labels.csv`, `manifest.json`, `crops/`, `crops_manifest.csv`, `crops_train.csv` | Training/label provenance |
|
| 34 |
+
| `decisive_lift_table.csv`, `lift_table.csv`, `lift_table_vs_real_photo.csv`, `per_class_metrics.csv`, `error_report.md` | Evaluation and augmentation-lift measurements |
|
| 35 |
+
|
| 36 |
+
## Headline results
|
| 37 |
+
|
| 38 |
+
- Detection: YOLO11n, mAP50-95 **0.76**; basket-hardened retrain: real 5-product basket 0 -> 7 boxes.
|
| 39 |
+
- Recognition: 25-way linear head on frozen DINOv2, **94%** validation (crop-aware).
|
| 40 |
+
- Serving (`head.onnx`/`head.int8.onnx`): **76.5%** on 17 held-out real photos, **96.2%** on held-out detector crops; ONNX parity 1.4e-6; int8 = 10.6 KB.
|
| 41 |
+
|
| 42 |
+
## License / usage restriction
|
| 43 |
+
|
| 44 |
+
**Educational use only.** Product studio images/prices originate from FairPrice/NTUC, collected
|
| 45 |
+
under fair-dealing/educational use for coursework — **do not use this repo or its contents
|
| 46 |
+
commercially or redistribute the dataset.** See the linked LICENSE for full terms.
|