Serve the v2 model: fetch meta.json with the backbone, size the GPU reservation for ViT-B

Serve the v2 model: fetch meta.json with the backbone, size the GPU reservation for ViT-B

Widen the segmentation reservation margin after measuring a cold start

Ask ZeroGPU for realistic durations so signed-out visitors can run

Cut the README down to what a maintainer needs

Replace em and en dashes with ordinary punctuation

Hide the Diagnostics panel from visitors

Use the description verbatim, no emphasis, no lead-in

Rewrite the page description

Drop the Notes block from the page

Retitle to match the model card; expand the description

Keep every classifier out of the main process (ZeroGPU abort fix)

Add CPU-side environment report

Fix ZeroGPU abort: build GPU models lazily, touch no CUDA at import

Diagnostics: test 60s vs 300s reservation

Add step-by-step ZeroGPU diagnostics

Fix: File input preserves the TIFF (Image re-encoded to PNG); raise ZeroGPU cold-start duration

RootScope Gradio app (ZeroGPU)

initial commit