Next-gen models: YOLO seg + SAM 2 refinement + experimental tooling

Custom orthogonalization (drop orthogonalizer pkg, fixes Python 3.13 build)

Add polygon simplification + orthogonalization for clean building footprints

RAMP: 4-channel watershed post-processing for adjacent building separation

Fix RAMP input normalization (divide by 255), add output debug logging

Switch tflite-runtime to ai-edge-litert (Python 3.13 compatible)

Add RAMP detection mode (aerial buildings specialist)

Expose Grounding DINO threshold as UI slider (0.05-0.50, default 0.15)

Fix mask fragmentation + add filter-stage logging

Heavy diagnostic logging for two-stage and single-stage modes

Fix Grounding DINO post-process kwarg: threshold instead of box_threshold

Add Grounding DINO + SAM 3 two-stage detection mode

Redirect deployment pipeline to duplicated el-best007 Space

Fix HF authentication URL to use universal 'user' syntax

Fix HF authentication URL mapping to use el-best007 instead of collzibip

Update GitHub Actions Token reference and test Auto-Sync

Add GitHub Actions workflow to auto-sync with Hugging Face Space

Add script to test programmatic agent interaction

Restore Building and add standalone Rooftop option

Merge pull request #1 from collin-schreyer/SAM3_BA

Update feature presets to test different building prompts

Show download files from first tile, update every 10 tiles

Pin PyTorch 2.8.0+cu129 to match driver CUDA 12.9 (was building with CUDA 13.0)

Fix GPU: single @spaces.GPU on segment_tile only, no nested decorators

Add checkpoint saves every 100 tiles with downloadable files mid-run

Fix GPU: move @spaces.GPU to segment_tile and load_model instead of generator

Add scan progress box to live preview, update every 10 tiles

Add GPU debug logging to diagnose CUDA detection issue

Add @spaces.GPU decorator to enable GPU allocation on HF Spaces

Fix GPU detection: use runtime check instead of import-time

Add 256px tile size option and set as default

Add tiling pipeline with live dashboard and confidence tracking

Add multi-feature support with presets and confidence scores

Add torchvision dependency for Sam3ImageProcessor

Replace GDAL with rasterio for GeoTIFF conversion

Fix Gradio SDK version to 5.29.0

Fix HF metadata: valid emoji and colorTo

Add Janus building footprint extraction app

initial commit