Spaces:
Runtime error
Runtime error
Commit History
-
fix(imports): import ClassUpdate and SubjectUpdate in api modules
d9f3f61 -
feat: enhance face mesh with decorative neural wireframe and harden admin master login
312354d -
feat(classes): add full edit and delete support for classes and curriculum subjects with interactive modals
ae2c4e1 -
feat: production readiness, admin master password protection, teacher registration approval gate, and data integrity safeguards
544faf7 -
feat: implement 1-identity = 1-face constraint with competitive bipartite matching
25f4c1c -
fix: resolve 405 Method Not Allowed and duplicate already recorded session with multi-protocol fallback
e6b4454
-
fix: resolve Method Not Allowed on attendance edit — add PATCH routes and switch frontend to PATCH
8adb1ca -
feat: collapsible sidebar, admin portal parallel sync with theme toggle, and resilient session record edits
a9d777c -
fix: resilient Excel export, rename to Mark Attendance, safe teacher_id session update, and review empty state
4552b75 -
feat: enhance mobile UI responsiveness with slide-out navigation drawer and touch-friendly controls
c4b299f -
fix(review): remove hardcoded mock faces fallback and tune YuNet group detection threshold
e854a64 -
feat: enable site-wide light and dark mode theme switching across all pages and components
68b71d4 -
fix: correct bounding box alignment offset calculation on image canvas
420d387 -
fix: resolve all 5 facial matching root causes
RC#1 embedder.py: 3-tier alignment fallback (prevents ~0.15 score drop from unaligned Haar fallback) RC#2 .env: Recalibrate thresholds for classroom conditions (match 0.45->0.38, margin 0.08->0.05) RC#3 attendance.py: Fix active-learning corruption - re-run YuNet before saving embeddings RC#4 quality.py+pipeline.py: classroom_mode=True flag removes false area-ratio rejections RC#5 imageCompressor.ts: quality 0.85->0.92 + maxDim 1920->2048 to preserve face edge detail5a777f0 -
feat: add AERON-inspired architectural logo, fix Excel export authentication, unify obsidian dark theme across all subpages, improve active navigation matching
634fbf6
-
fix: resolve BoundingBox w/h and width/height typescript error for Vercel build
d302462 -
fix: resolve BoundingBox width and height property type error for Vercel deployment
52f9301 -
feat: add HeroLanding page, clean camera overlay, fix seed safety guard, fix TS build error
0f7caca -
fix: attendance history upsert, auto-refresh history page, multi-photo active learning
- Backend: Replace 409 duplicate error with UPSERT logic (delete+recreate records on re-submission) - Backend: Fix online active learning for multi-photo sessions (use image_urls array + image_index) - Backend: Fix history GET total_enrolled fallback for accurate 0-count display - Schema: Add image_urls field to SaveAttendanceSessionRequest for multi-photo support - Frontend: History page auto-refreshes on navigation (useLocation.key dependency) - Frontend: Add manual Refresh button to History page - Frontend: Color-coded attendance percentage badges (green/amber/red) - Frontend: Pass image_urls array when saving session for full multi-photo learning4c070cc -
fix: add missing AttendanceAuditLog import causing save crash
5bfee0d -
fix: face recognition accuracy - consistent embeddings, lower threshold, EXIF orientation
- Force all embedding extraction through OpenCV FaceRecognizerSF.feature() instead of broken ONNX Runtime path that had wrong normalization (raw 0-255 pixels, no mean subtraction) - Lower YuNet face score threshold from 0.4 to 0.25 to detect more classroom faces (tilted, glasses, dim) - Haar cascade fallback now re-runs YuNet on cropped faces to extract 5-point landmarks for proper alignment - Add EXIF orientation correction for mobile photos that arrive rotated sideways432d1c8 -
fix: multi-image face detection crash - NameError, duplicate upload, error handling
- Fix NameError in pipeline.py: max_workers was never defined in process_multiple_classroom_images, causing crash on every multi-photo analysis - Fix frontend sending first photo twice (as both 'files' and 'file'), which triggered the multi-image path even for single photos - Fix backend file dedup: use 'file' param only as fallback when 'files' is empty - Broaden exception handling in attendance API to catch all pipeline errors, not just ValueError197ec85
-
fix(cv): eliminate OpenCV C++ thread collision in multi-photo analysis
4ed6bcf -
fix(api): standard FastAPI Form and File parameters for analyze endpoint
2cb8ca5 -
fix(api): ensure 100% universal file stream extraction from form data
3f28e23 -
fix(storage): use save_bytes to prevent stream EOF corruption on file uploads
8102798 -
fix(api): remove hardcoded multipart Content-Type header and refine multi-image file list extraction
c4626db -
fix(core): resolve React Minified Error #31 root cause and eliminate all raw object error rendering
a7d05db -
feat(cv): parallel multi-photo processing with ThreadPoolExecutor and cross-image deduplication
ec88759 -
feat(attendance): add multi-image upload & multi-shot webcam support for whole-classroom coverage
02e4fcf -
fix(db): add connection test and auto-fallback to SQLite when remote DB is unreachable
09c6a5b -
fix(cv): add missing Dict import in FaceEmbedder
f6df62f -
feat(cv): add ONNX Runtime acceleration, batch matching, and 3D Neural Mesh scanning overlay
c754de4
-
feat(admin): aesthetic dedicated admin portal at /admin protected by master password 2026/
bb13d79 -
feat: continuous face training studio and remove sample data
40d9f19 -
feat(api): native FastAPI routing with Gradio mounted at /gradio
a18e7ec -
feat(auth): add self-service account creation and student portal switch
f2507ad -
fix(gradio): patch schema parser for pydantic v2 and add reliable model mirrors
485e117 -
feat: switch to Docker SDK for reliable HF Spaces deployment
Root cause: With sdk:gradio, the HF supervisor expected demo.launch() but our app.py only defined variables and exited — causing restart loops. Port 7861 bind conflicts occurred on successive restarts. Fix: - Switch to sdk:docker with a proper Dockerfile (full server control) - app.py now starts uvicorn on port 7860 via __main__ - Auto-seeds database on first boot if no users exist - Removed gradio dependency (saves ~500MB+ in Docker image) - Added HTML landing page at / with links to Swagger docs - Added .dockerignore for faster builds30dfbaf -
fix: add missing Optional typing import in excel_service.py
e4efb6a -
fix: robust database URL parsing and safe fallback
ebee7b7 -
fix: resolve app module namespace in Hugging Face entrypoint
e248a74 -
feat: complete Attendx smart attendance system with facial recognition
67ac4d2