Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,90 @@
+---
+license: apache-2.0
+tags:
+- computer-vision
+- saliency-detection
+- poster-generation
+- evaluation
+- isnet
+- basnet
+---
+
+# PosterO Saliency Detection Models
+
+This repository contains the saliency detection model weights for the PosterO evaluation pipeline.
+
+## Models Included
+
+### ISNet (isnet-general-use.pth)
+- **Purpose**: High-quality saliency detection
+- **Architecture**: ISNetDIS
+- **Input Size**: 1024×1024
+- **Usage**: Primary saliency map generation
+
+### BASNet (gdi-basnet.pth)  
+- **Purpose**: Complementary saliency detection
+- **Architecture**: BASNet
+- **Input Size**: 256×256
+- **Usage**: Secondary saliency map generation
+
+## Usage
+
+These models are automatically downloaded and used by the PosterO evaluation script:
+
+```bash
+# Basic usage - models download automatically
+python run_saliency_and_eval_hf.py \
+    --input_dir ./images \
+    --predictions ./predictions.json \
+    --output_dir ./results
+
+# Or specify the repository explicitly
+python run_saliency_and_eval_hf.py \
+    --input_dir ./images \
+    --predictions ./predictions.json \
+    --output_dir ./results \
+    --hf_isnet_repo "pengdaica/saliency_weights" \
+    --hf_basnet_repo "pengdaica/saliency_weights"
+```
+
+## Evaluation Pipeline
+
+The final saliency map used in evaluation is computed as the **element-wise maximum** of both models:
+
+```python
+final_saliency = np.maximum(isnet_map, basnet_map)
+```
+
+This approach leverages the strengths of both architectures:
+- ISNet provides high-resolution, detailed saliency detection
+- BASNet offers complementary detection patterns
+- Maximum operation captures the union of salient regions
+
+## Model Details
+
+| Model | File | Size | Resolution | Framework |
+|-------|------|------|------------|-----------|
+| ISNet | `isnet-general-use.pth` | ~168 MB | 1024×1024 | PyTorch |
+| BASNet | `gdi-basnet.pth` | ~332 MB | 256×256 | PyTorch |
+
+## Installation
+
+```bash
+pip install torch torchvision huggingface_hub pillow opencv-python numpy matplotlib cairosvg tqdm
+```
+
+## Citation
+
+If you use these models, please cite the original PosterO paper and the respective saliency detection methods:
+
+- ISNet: [Intermediate Supervision Network for Salient Object Detection](https://arxiv.org/abs/2109.12172)
+- BASNet: [Boundary-Aware Segmentation Network for Mobile and Web Applications](https://arxiv.org/abs/2101.04704)
+
+## License
+
+Apache 2.0
+
+## Repository
+
+- **Code**: [PosterO-CVPR2025](https://github.com/your-repo/PosterO-CVPR2025)
+- **Models**: [pengdaica/saliency_weights](https://huggingface.co/pengdaica/saliency_weights)