Relax AI rejection thresholds and add density guardrail

Harden AI rejection logic and sync specialist disease database

Harden rejection thresholds and cleanup

Hardened non-plant rejection logic (35% threshold)

Fix build error: standardized icon dimensions to 1:1 square

Updated app branding with professional logo icon

Finalized accuracy fixes and linked to production Hugging Face backend

Upgrade accuracy engine to Multi-Expert (ViT Anchor) and Unify Dashboard

Filter specialist by generalist plant species

Fix Syntax Error

Fix Indentation Error

Fix Indentation Error

Aggressive Specialist crop override logic

Aggressive Specialist crop override logic

Fix dictionary mapped strings for Specialist classes

Fix BGR to RGB color conversion for Raspberry Pi camera feed

Fix pixel normalization hallucination

Fix pixel normalization hallucination

Force HuggingFace build trigger

Force HuggingFace build trigger

Aggressive float coercion

Aggressive float coercion

Fix tensor precision mismatch

Fix tensor precision mismatch

Add debug statements

Add debug statements

Eliminate torchvision requirement completely

Eliminate torchvision requirement completely

Add torchvision to fix AutoImageProcessor

Add torchvision to fix AutoImageProcessor

Physical pipe bypass

Physical pipe bypass

Fix image processor crash

Fix image processor crash

Fix specialist string map keys

Fix specialist string map keys

Completely disable Gen from overriding Spec on Tomato

Completely disable Gen from overriding Spec on Tomato

Fix collision logic

Fix collision logic

Fix collision between Tomato and Potato hallucination

Fix collision between Tomato and Potato hallucination

Fix AI engine confidence dropping Rice Leaf Blast

Restored versatile detection with 3-expert model and relaxed focalizer

Update Hugging Face Space with latest backend and deployment config

RESTORE PRECISE DETECTION: Specialist-first logic

Fix AI detection accuracy and ensemble logic

Fix backend torch version and APK connectivity issues

Add real-time Raspberry Pi detection with Streamlit dashboard

Allow lower confidence for uncertain images