14 lines (11 loc) · 541 Bytes

Future Roadmap & TODOs

Immediate Priorities

Add support for Audio retrieval (using AudioCLIP).
Optimize FAISS index for >1 Million images (IVF_PQ).

Long-term Goals

Deploy as a Docker container for easier distribution.
Implement "Re-Ranking" mechanism after retrieval to improve accuracy.
Upgrade BLIP-2 to LLaVA or GPT-4o API integration.

Known Limitations

Current inference speed is ~70s on RTX 3060 (Need optimization via TensorRT).
LoRA Fine-tuning limited to domain-specific captions only.