---
title: XDL Colitis Demo
emoji: 👁
colorFrom: pink
colorTo: red
sdk: gradio
sdk_version: "6.5.1"
app_file: app.py
pinned: false
---

# XDL Colitis Demo

Simple Gradio demo for DenseNet-based colitis image classification with optional XDL overlays.

## Features

- Batch inference only
- Case selector:
  - Multiclass (4 Classes)
  - Crohn vs TB
  - Kolitis Ulseratif vs Infeksi
- Folder upload support (`gr.File` directory upload)
- Optional local folder-path input
- Automatic device detection (`cuda` if available, else `cpu`)
- Presentable summary card for key outputs (device, processed count, final class, mean confidence, etc.)
- Final folder classification by majority vote of per-image top class
- Mean confidence for the selected final class
- Compact GradCAM + SmoothGrad overlays for classified images
- Optional saving of compact XDL results to a local folder

## Run

```bash
.venv/bin/python app.py
```

Then open the Gradio URL in your browser.

## Project Structure

- `app.py`: minimal launcher
- `src/ui.py`: batch-only Gradio UI
- `src/inference.py`: model loading + inference + batch/XDL pipeline
- `src/xdl.py`: DenseNet-specific XDL helper functions
- `src/densenet.py`: model definition
- `src/preprocessing.py`: inference transform

## Notes

- Default model paths:
  - `models/multiclass.pth` (Multiclass)
  - `models/crohn_tb.pth` (Crohn vs TB)
  - `models/ku_infeksi.pth` (Kolitis Ulseratif vs Infeksi)
- You can override model paths with env vars:
  - `MODEL_PATH_MULTICLASS`
  - `MODEL_PATH_CROHN_TB`
  - `MODEL_PATH_KU_INFEKSI`
- If model loading fails, the UI returns a clear error instead of silently using random weights.