Nikhil3235/fire-smoke-detection

Upload upload_log.txt with huggingface_hub

Upload static/sample-smoke.mp4 with huggingface_hub

Upload static/sample-fire.mp4 with huggingface_hub

Upload models/best.pt with huggingface_hub

Update README.md with premium layout, system architecture, mathematical anti-spoofing formulations, and local setup guide

\Fix static image upload rendering by adding raw image reading and processing inside generate_frames"

Fix video feed lag by using fast camera.grab() frame skipping, fix missing bounding boxes on static video files by adding draw_boxes=True parameter, and fix stop siren race condition on stats fetch

Fix critical bracket syntax error in script.js to restore webpage functionality and enable camera start

Fix webcam failure by adding camera getUserMedia fallback constraints when facingMode environment is not supported on desktop devices

Fix auto-saves to trigger only above 55% hazard confidence, clear alert gallery UI on monitoring stop, and implement static standard deviation tracking filter for printed photo frames

Optimize person model with imgsz=640 and conf=0.25 to catch humans instantly with 100% precision. Reset stats on stream stop. Implement persistent history.json safety incident database on server with UI table logging

Optimize video file playback to 1.0x real-time speed using elapsed wall-clock frame skipping, add progress logging to uploads via XMLHttpRequest, and bump script cache version to v=16

Optimize cold-start delay by adding startup YOLO model warmups with dummy frame inference, and bump cache version to v=15

Restructure site with separate /live workspace, fix human/shirt false fire positives, remove pitch-black ghost detections, and implement 60 FPS LERP animation for bounding boxes with zero latency ping-pong frame submission

Optimize dual models inference to run at imgsz=320 fixing the 0.1 FPS lag issue caused by CPU overload. Fixed YOLO hallucination bug on pitch black frames by raising Person confidence threshold to 0.60

Implement anti-spoofing screen detection to ignore videos of fire on mobile phones/TVs. Added Fake Fire (Screen) UI logic in JS to draw yellow boxes without triggering alerts.

Fix UI bugs: remove 8500% bug by removing multiplication by 100, remove percent sign from Person label, and lower Doll threshold to 0.45 to prevent false dolls

Implement dual AI models (fire/smoke + person detection). Added living vs doll heuristic. Increased fps to 20 with sequence numbering to fix network out-of-order latency bugs causing sticky alarms and stuck boxes. Raised fire conf threshold to 65% to ignore sun/lights.

Fix mobile portrait camera distortion causing YOLO to miss smoke, and fix server queue lag by restoring isProcessingFrame lock with 8fps target

Implement Session Isolation & Ultra-Low Latency (<0.1s) concurrent frame processing

Optimize webcam capture loop: Increase framerate to 12fps and use full 640x360 resolution for hyper-fast small fire detection

Add mobile-friendly camera flip toggle to switch between front and rear cameras

Implement beautiful dynamic glowing flame cursor particle trail and bump version to v6

Optimize webcam stream with real-time local video rendering, local canvas overlay, and optimized JSON bounding box transmission

Fix video autoplay blocks and remove metadata Promise lock to prevent black screen webcam hang

Add checkCameraAvailability banner diagnostic info and bump version to v3

Fix silent JavaScript TypeError crash by adding safety guards for non-existent statusDot and statusText elements

Force browser cache bust for style.css and script.js

Add camera availability check and user-facing alert popup for diagnostic logging

Implement true client-side live webcam streaming via /api/process_frame and fix premium toggle styling and profile shimmer glow

Remove models and videos from repository .gitignore to fix Hugging Face API upload ignore bug

Add .gitattributes to track sample videos and model weights in Git LFS

Log asset uploader output and upload log file to space for debugging

Enhance debug_status endpoint to list root, models, and static directories

Update upload_assets.py to fail on error for traceback debugging

Add asset uploader to workflow to bypass runtime firewall restrictions on Hugging Face Spaces

Add debug_status endpoint for remote server diagnosis

Implement dynamic sample video download and auto-fallback when webcam is unavailable on cloud server

Update model download URL in app.py to Nikhil's custom-trained best.pt model

Fix permission error by creating models directory and chowning /app to user in Dockerfile

Use Debian Bookworm base image in Dockerfile to resolve libgl1-mesa-glx dependency error

Fix Hugging Face colorTo metadata validation error by using yellow

Add Hugging Face Space configuration metadata to README.md

Initial commit: Intelligent Fire & Smoke Detection system by Nikhil Mali