Born at hackathon, not built for it

Update banner image

Update README: fix training data, add banner and video

Add banner image

CSS: responsive table for mobile

fix: bold only best AUC per pathology in results table

Triple image resize: webcam + client upload + server fallback

Fix client-side resize: intercept fetch + limit webcam resolution

Fix example3_effusion - only Effusion detected

Client-side image resize for mobile photos

Optimize example5_multiple.jpg for mobile

Optimize example4_edema.jpg for mobile

Optimize example3_effusion.jpg for mobile

Optimize example2_cardiomegaly.jpg for mobile

Optimize example1_normal.jpg for mobile

Fix example4_edema - only Edema detected

Fix example2_cardiomegaly

Fix example1_normal

Fix torch_dtype deprecation, remove scroll JS

Fix scroll using parentIFrame API for HF Spaces

Fix scroll with retry on example click

Fix example scroll with capture phase

Fix scroll to image on example click

Force rear camera via getUserMedia override

Fix rear camera and responsive cards

Fix mobile: responsive cards, rear camera, scroll

Add expected results table, rear camera support, scroll to top on example click

Replace example1 with patient64775 (true negative)

Replace example1 with patient64765 (true negative)

Add 5 CheXpert example X-rays with gr.Examples

Upload examples/example3_effusion.jpg with huggingface_hub

Upload examples/example1_normal.jpg with huggingface_hub

Upload examples/example5_multiple.jpg with huggingface_hub

Upload examples/example2_cardiomegaly.jpg with huggingface_hub

Upload examples/example4_edema.jpg with huggingface_hub

Cache model on GPU to avoid CPU->GPU transfer each request

Fix ZeroGPU: move model to cuda inside @spaces.GPU

Fix: load fp16 at module level for ZeroGPU compatibility

Add torchvision (required by Gemma4VideoProcessor)

Load merged model directly (no PEFT needed)

Use transformers+peft instead of unsloth for inference

Replace unsloth with transformers+peft for inference (no GPU needed at import)

Try gradio 5.25.0 to fix gradio_client schema bug

Fix: replace gr.JSON with gr.Textbox to avoid gradio 5.x schema bug

Upgrade to gradio 5.12.0 (compat with huggingface_hub>=0.30)

Simplify requirements (Python 3.10 has audioop natively)

Force Python 3.10 to fix gradio/unsloth/audioop compat

Fix: audioop-lts (not pyaudioop) for Python 3.13 pydub compat

Fix: match gradio==4.44.0 (HF enforced) + pyaudioop for py3.13

Fix: pin gradio>=5.0 + add pyaudioop for Python 3.13