Swap stage-2 v3 → v4 (mAP@.5 0.572 → 0.601; caries 0.035 → 0.071 from DENTEX-c diagnosis labels at 1664x928)

AI Radiology Report: drop 404 model fabrications, add user-chosen 5-model list, throttle yields to ~14/sec for smoother streaming

AI Radiology Report: stream tokens progressively (~2s to first visible token vs 8s+ for full response)

Assignment metric: containment → IoMin (fixes RC-treated false-unassigned when bbox larger than tooth)

Report tab: rename to "AI Radiology Report" + use generator for reliable progress feedback

Report tab: visible "Generating…" feedback + button disable while Ollama call in flight

Per-tooth tab: only annotate teeth WITH findings + adult-OPG tooth-bud suppression

Fix dissertation-inherited assignment metric (IoU→containment, 0.5 threshold, best-fit wins)

Phase 8: add Medical report tab (Ollama Cloud + Gemma 4 / Qwen / DeepSeek / Kimi)

Phase 8: add Medical report tab (Ollama Cloud + Gemma 4 / Qwen / DeepSeek / Kimi)

Revert stage-1 v4 → v3 (per user feedback after side-by-side check)

Swap stage-1 v3 → v4 (mAP@.5 0.853 → 0.862; FDI 48 0.41 → 0.56 via archive merge)

Phase 5d: add per-tooth findings tab (IoU-match stage-1 ⨯ stage-2 outputs)

Switch stage-2 to v3 (cold-start); threshold 0.005 to match score range

Extend demo to two-stage: stage-1 v3 (FDI) + stage-2 v2 (conditions), tabs + dual sliders

Switch to stage-1 v3 (DENTEX-augmented); threshold 0.02 to match v3 score range

Switch to stage-1 v2; add per-class top-1 dedup + threshold 0.05 + honest banner

Bump gradio to 5.34.0 (fix get_api_info bool-additionalProperties bug)

Initial deploy: README with Space frontmatter

Initial deploy: requirements

Initial deploy: stage-1 FDI tooth detector demo

initial commit