SuhxsReddy/SingaporeAnalytics

fix(network): head-on camera y-anchors + signpost filter (v7)

feat(network): anchor-based N-direction visibility (v6) + dataset v2

feat(network): use ground-truth camera_config.json for direction labels (v5)

Add camera analysis notebook — ground truth config from LTA text labels

Build network before first sweep — direction labels correct from camera 1/90

Fix: network build no longer blocks inference loop

Proper N-direction counting via visibility analysis

Add per-lane direction counts to dataset

Simplify dataset: lane detection internal only, dataset stays 2-directional

Add per-lane detection + lower IOU for articulated vehicles

All remaining refinements before overnight collection run

Rebuild camera network properly — all four fixes

Add camera road network with OCR-based direction labelling

Schema validation on startup — detect and discard stale CSV format

Add conf/iou/imgsz/model_version columns to dataset for versioning

Increase imgsz to 1280 for better distant/small vehicle detection

Proper directional counting via 2-frame IoU tracking + lower IOU threshold

Log dataset push errors to header instead of silently swallowing them

Fix: set neck_channels=[128,256,512] for Phase 2 CATI weights

Fix thread startup and add pandas to Dockerfile

Fix: load model in main Streamlit thread before passing to inference thread

Save annotated detection images from first sweep to HF Dataset

Add HF Dataset persistence — resume data across Space restarts

Redesign app as continuous analytics dashboard

Fix transparent overlay: disable cache spinner, simplify detection tab layout

Fix detection image rendering — replace preview in-place using st.empty()

Lower default confidence to 0.10, add conf slider and zero-detection hint

Add CATI detection tab and Phase 2 inference pipeline

Unpin Dockerfile deps to fix HF build error

Fix HF short_description length

CATI Singapore Smart City Analytics — initial deploy