|
|
--- |
|
|
title: MMRM |
|
|
emoji: 🐠 |
|
|
colorFrom: yellow |
|
|
colorTo: blue |
|
|
sdk: gradio |
|
|
sdk_version: 6.5.1 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: gpl-3.0 |
|
|
short_description: 'Restoring Ancient Ideograph: A Multimodal Multitask Neural N' |
|
|
--- |
|
|
|
|
|
|
|
|
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference |
|
|
|
|
|
## Interactive Demo |
|
|
|
|
|
This Gradio app demonstrates the restoration capabilities of the MMRM model compared to textual and visual baselines on real-world damaged character data. |
|
|
|
|
|
### Features |
|
|
- **Real-world Data**: Select from samples in the `data/real` directory. |
|
|
- **Model Comparison**: |
|
|
- **Zero-shot Baseline**: Pre-trained GuwenBERT (Works out-of-the-box without training). |
|
|
- **Textual Baseline**: Fine-tuned RoBERTa. |
|
|
- **Visual Baseline**: ResNet50. |
|
|
- **MMRM**: Our proposed Multimodal Multitask Restoring Model. |
|
|
- **Intermediate Visualization**: Shows the restored image generated by the MMRM capability. |
|
|
|
|
|
### Running the Demo |
|
|
|
|
|
1. **Deploy to Hugging Face Spaces**: |
|
|
- Create a new Space on Hugging Face (SDK: Gradio). |
|
|
- Upload the contents of this `demo` folder to the Space repository. |
|
|
- Upload your model checkpoints to the `checkpoints/` folder in the Space. |
|
|
- `checkpoints/phase2_mmrm_best.pt` |
|
|
- `checkpoints/phase1_roberta_finetuned.pt` |
|
|
- `checkpoints/baseline_img.pt` |
|
|
|
|
|
*Note: Even without checkpoints, the demo will run using the Zero-shot Baseline (downloaded automatically).* |
|
|
|
|
|
2. **Local Testing**: |
|
|
- Install requirements: `pip install -r requirements.txt` |
|
|
- Run: `python app.py` (assuming you are inside the `demo` directory) |
|
|
|
|
|
|
