Fix: add einops dependency required by AdaptFormer custom code

Fix: add trust_remote_code=True for AdaptFormer custom model code

Production overhaul: pre-trained AdaptFormer model + detection quality improvements

Optimize notebook for CPU training: MobileNetV2 encoder, 15 epochs, batch 4

Fix notebook: use HuggingFace datasets CDN instead of flaky Google Drive links

Add Siamese U-Net training notebook and model inference integration

Clean visualization: subtle tint overlay, outline-only boxes, 30-region cap, high-contrast labels

Redesign detection pipeline: CVA, vegetation/shadow suppression, SNR fusion, multi-scale SSIM

Thinner, more precise detection lines and compact labels

Restore Method dropdown, Image Registration and Radiometric Normalization options

Force cache bust - bump APP_BUILD, CSS and JS versions

Remove method/model selection UI - use AI-Based Deep Learning by default

Revert to satellite change detection only - remove landslide, pothole, detection type menu

Pothole YOLO pipeline: single-image mode, modular detector, red box visualization

Add pothole detection as new menu type with separate engine and research plan

HF runtime fix: bind uvicorn to dynamic PORT env

UX: always land on detection-type menu after login

Fix detection-type menu navigation: switch mode and dashboard without reload loop

UX: show detection-type menu after auth then auto-redirect based on URL

HF: force writable hub cache (/tmp/hf)

HF: relax pip wheel-only; upgrade pip tools

HF: ensure pip uses binary wheels during build

HF: make Docker builds verbose; move APP_BUILD cache-bust before pip

UX: show detection-type menu after login before redirect

HF: bump APP_BUILD to force rebuild

UX: add choose detection type screen and redirect URLs

HF: bump APP_BUILD to force clean Docker rebuild

Add landslide detection menu, separate engine, and Uttarakhand integration plan

AI/Hybrid overhaul: bidirectional AI fusion, stronger hybrid sensitivity, keep large unclassified regions

AI rework: percentile+blur threshold + less aggressive cleanup; Hybrid lower threshold

Fix: fallback to Image Difference when AI/Hybrid mask is too sparse

Detection fix: ECC registration fallback + less aggressive large-region rejection

Detection tuning controls + threshold diagnostics in API response

UI: show pseudo progress percentage while analyzing images

Email API: attach populated ChangeDetection.html report (base64)

Email API: send required FileName and AttachmentBase64 placeholders

Use manager email API for notifications (multipart/form-data); SMTP fallback

Revert Resend; use SMTP only, custom email API to be integrated later

Use Resend HTTPS API for email (HF Spaces); SMTP fallback for local

Force fresh HF deploy with version 2.2.0 and new asset cache keys

Make HF startup path simpler and add explicit startup logging

Rework email notifications with explicit send actions.

Switch email from SMTP_SSL:465 to STARTTLS:587, restore sender, show failure feedback

Add pagination to regions table in result modal (10 rows per page)

Force Docker rebuild with cache-bust ENV to deploy IST timestamp fix

Bump version to 2.1.0, add version+IST time to /health for deploy verification

Convert timestamps to IST server-side so display is correct regardless of browser cache

Force history timestamps to IST (GMT+5:30 Asia/Kolkata)

Fix history timestamps: append UTC offset so browser converts to local time

Full codebase audit: fix critical perf bug, security, error handling, dead code