Product Detection Pipeline

Click any block to see details and the command that runs it.

Data Models Scripts Deploy Manual step
Shelf images
jpg / png / mp4 / webcam
YOLOv8n detector
Detect & crop every product
Path A — first run Path B — iterative
1_generate_crop_dataset.py
Crop only, no classifier
Manual labelling
Move crops to class folders
auto_label.py
Detect + auto-classify
Browser review UI
Confirm / fix / reject
data/
Labelled crops by class folder
1b_split_dataset.py
Stratified train / val / test
crops_dataset/
train / val / test
Local GPU / CPU Free cloud GPU
2_train_classifier.py
EfficientNet-B0 local
Kaggle notebook
T4 GPU — free tier
Classifier weights
best.pt + class_names.json
3_inference.py
Images · video · webcam
api_server.py
REST + polling API
↻  Iterative loop — shoot new shelf photos → run auto_label.py → review → retrain → better auto-labelling next time