revert: restore original single-video output with baked annotations

feat: separate clean video from annotated video for interactive playback

fix: resolve backend-frontend integration issues across API contract, error handling, and resource management

refactor: remove depth estimation pipeline entirely

fix: simplify review — fix mask NMS bug, remove dead code, hoist imports

feat: add universal post-detection NMS before tracking

feat: hide class labels on non-mission-relevant objects in rendered video

feat: mission-relevance coloring — red for relevant objects, gray for non-relevant

fix: clean up video display — remove track ID labels, smaller fonts, masks-only for segmentation

feat(inspection): persist RLE masks during segmentation inference

feat(isr): store latest frame in writer thread for ISR cropping

feat: add TTFS instrumentation to run_inference pipeline

fix: steady 24fps MJPEG streaming with no frame drops

fix: weight download race condition + rewrite CLAUDE.md

refactor: remove first-frame tab and simplify to video-only UI

refactor: strip threat fields from all frontend and backend code

feat: add GPT vision-based mission relevance assessment

fix: hide instance_id from segmentation overlay labels

Eliminate redundant JPEG frame loading via shared frame store

feat(ttfs): instrument segmentation time-to-first-stream

refactor: rename hf_yolov8 → yolo11 across codebase

chore: remove dead code and unused frontend modules

refactor(gsam2): make SAM2 detector-agnostic

fix: replace gated GSAM2 streaming with adaptive-rate publisher thread

perf: defer GPU->CPU mask transfer to render thread via LazyFrameObjects

fix: correct benchmark metrics — remove double-counting and add missing timers

perf: pipeline GSAM2 tracking + rendering with startup buffer

removed extra bbox lable and track id

feat: add num_maskmem parameter to /benchmark endpoint

perf: GPU-resident tensor pipeline for SAM2 video propagation

feat: add detailed timing metrics to GroundedSAM2 segmenter

feat: add job processing timing and latency measurements to backend and frontend

feat(backend): enhance inference pipeline with GLM logic and structured outputs

Suppress labels for non-relevant objects in visualization

Fix GSAM2 label rendering double-draw and sync GPT enrichment to job storage

Handle GPT refusal paths and preserve assessment status

Fix segmentation mission routing and first-frame GPT alignment

Fix GSAM2 GPT writer state safety and background call args

Fix cross-GPU mask device mismatch in GSAM2 reconciliation

Harden GSAM2 parallel pipeline and tracking reconciliation

Refine Grounded-SAM2 tracking behavior and video frame handling

Remove SAM3 and standardize segmentation on Grounded-SAM2

Fix Grounding DINO load reliability and writer error handling

Add diagnostic logging to pinpoint tracker dropping detections

Fix GPT thread safety, improve relevance logic, and add caching for COCO matching

Async GPT enrichment: fire-and-forget background task updates track cards from UNASSESSED to ASSESSED

Simplify first-frame processing, replace depth pre-scan with incremental stats, add GPT enrichment thread

Add StreamingVideoWriter (ffmpeg pipe), reuse first-frame GPT results in inference pipeline

Update overlays, tracker, GPT reasoning, and threat chat improvements

feat: Add LLM post-processing relevance gate for broad detection