Disable NV VO-boundary masking by default

Adjust VO default threshold

Add PMC full-text ingestion, fix inference SVG gradients, remove header mouse toggle

Brighten widget labels and style tabs for dark theme

Fix metric denominator, st.stop scope, and deduplicate source rendering

Collapse sidebar by default and enable TTA by default

Update default thresholds: VO → 0.99, BG → 0.63

Fix relative paths for checkpoint — resolve from project root

Track ChromaDB binary files with Git LFS

Prepare for Hugging Face Spaces deployment

Defer RAG import to prevent OOM during segmentation inference

Force CPU-only torch on Streamlit Cloud to fix OOM/buffering

Add uv.lock to .gitignore — platform-specific lock breaks cross-OS deploys

Remove uv.lock — Windows-locked torch wheels break Linux deploy

Revert torch CPU suffix — Streamlit Cloud uses uv which resolves it automatically

Ship pre-populated ChromaDB (977 abstracts) and fix torch install on Streamlit Cloud

Added Dev Container Folder

Remove auto-ingestion from startup to fix deploy timeout

Cap batch processing at 50 images to stay within memory limits

Add uv.lock for reproducible dependency resolution

Track ChromaDB and auto-ingest PubMed abstracts on first deploy

Fix st.image deprecation: replace use_container_width with width="stretch"

Update HF repo ID to OIRSEG/OIRSEG_V1 and add huggingface-hub dependency

Prepare for HF Spaces deployment

Add CLAUDE.md with project conventions and domain knowledge

Remove Docker files — not needed for current deployment

Add Docker deployment (Dockerfile, docker-compose, .dockerignore)

Add ruff, pre-commit, and pytest

Add project presentation PDF

Initial commit: OIRseg retinal image segmentation app