How siddhartha-sai-17/product-vision-recommedation works
This repository is a fashion-image visual search and recommendation product, with a Streamlit dashboard and a CLI for preparing data, training models, and evaluation. The user workflow is image upload, preprocessing, embedding selection, similarity retrieval, metadata filtering/ranking, and product display. Other documented branches include model comparison, evaluation, dataset exploration, explainability, and live telemetry. Several page implementations are unsampled; those capabilities and their paths come from the README and tree, so uncertain internal links are omitted.
Main entry points: main.py Pipeline CLI · recommendation_service.py Recommendation service · 2_Image_Search.py Image search · cache_manager.py Model and index cache
Components 30
Dashboard
- Streamlit entry
app/streamlit_app.py - Image search
app/pages/2_Image_Search.py - Model comparison
app/pages/3_Model_Comparison.py - Evaluation dashboard
app/pages/5_Evaluation_Metrics.py - Dataset explorer
app/pages/7_Dataset_Explorer.py - Telemetry and logs
app/pages/8_System_Logs.py - Search session state
app/utils/session_manager.py - Performance dashboard
app/pages/4_Performance.py
Visual Retrieval
- Embedding selection
recommendation/embedding_selector.py - Recommendation entry
recommendation/recommendation_engine.py - Recommendation service
recommendation/recommendation_service.py - Similarity search
recommendation/similarity_engine.py - Filter and rank
recommendation/ranking_engine.py - Model and index cache
recommendation/cache_manager.py - Embedding database
recommendation/embedding_database.py - FAISS index
recommendation/faiss_engine.py
Data and Models
- Image preprocessing
preprocessing/image_preprocessor.py - Model architectures
models/model_factory.py - Model loader
models/model_loader.py - Model checkpoints
models/checkpoint_manager.py - Dataset preparation
preprocessing/dataset_subset.py - Dataset validation
preprocessing/dataset_validator.py - Transfer training
training/trainer.py - Siamese training
training/siamese_training.py - Pipeline CLI
main.py
Evaluation and Operations
- Retrieval evaluation
recommendation/evaluation_engine.py - Model comparison
recommendation/comparison.py - Reports
recommendation/report_generator.py - Plots and Grad-CAM
utils/visualization.py
Other components
- Dashboard user
Connections 34
- Dashboard user → Streamlit entry: opens
- Streamlit entry → Search session state: initializes state
app/streamlit_app.py - Dashboard user → Image search: uploads image
- Image search → Recommendation entry: requests recommendations
app/pages/2_Image_Search.py - Image search → Search session state: uses search state
app/pages/2_Image_Search.py - Recommendation entry → Recommendation service: delegates to
recommendation/recommendation_engine.py - Recommendation service → Image preprocessing: preprocesses query
recommendation/recommendation_service.py - Recommendation service → Embedding selection: selects embedding
recommendation/recommendation_service.py - Recommendation service → Model and index cache: loads cached assets
recommendation/recommendation_service.py - Recommendation service → Similarity search: computes similarity
recommendation/recommendation_service.py - Recommendation service → Filter and rank: ranks candidates
recommendation/recommendation_service.py - Model and index cache → Embedding database: loads vectors and metadata
recommendation/cache_manager.py - Model and index cache → FAISS index: loads or builds index
recommendation/cache_manager.py - Embedding selection → Model checkpoints: selects checkpoint
recommendation/embedding_selector.py - Model loader → Model architectures: creates model
models/model_loader.py - Model loader → Model checkpoints: loads checkpoint
models/model_loader.py - Dashboard user → Model comparison: compares models
- Dashboard user → Evaluation dashboard: views metrics
- Dashboard user → Dataset explorer: explores catalog
- Dashboard user → Telemetry and logs: views telemetry
- Dashboard user → Performance dashboard: views performance
- Pipeline CLI → Dataset preparation: prepares dataset
main.py - Pipeline CLI → Dataset validation: validates dataset
main.py - Pipeline CLI → Transfer training: starts training
main.py - Pipeline CLI → Siamese training: starts training
main.py - Transfer training → Model architectures: builds architecture
training/trainer.py - Transfer training → Model checkpoints: saves checkpoints
training/trainer.py - Siamese training → Model architectures: builds architecture
training/siamese_training.py - Pipeline CLI → Model comparison: runs comparison
main.py - Pipeline CLI → Reports: generates report
main.py - Dashboard user → Plots and Grad-CAM: views explainability
- Evaluation dashboard → Retrieval evaluation: presents metrics
- Model comparison → Model comparison: presents results
- Performance dashboard → Telemetry and logs: shares telemetry