Pyramid Feature Attention Network: CVPR 2019 Saliency Detection with DUTS

Sairajk developed this PyTorch implementation of the CVPR 2019 Pyramid Feature Attention Network for saliency detection, with the most recent update in July 2024. It provides the framework for training and inference on the DUTS dataset using edge-loss and attention mechanisms. The repository includes pretrained models and specialized dataloaders for salient object detection tasks.

Use Cases

• Training salient object detection models using the provided DUTS dataloader
• Generating saliency masks for custom images using the provided pretrained models
• Implementing edge-loss functions to refine object boundaries in segmentation tasks

Strengths

• Implements CVPR 2019 peer-reviewed architecture
• Includes pretrained models for immediate inference
• Features specialized edge-loss labels for boundary refinement

Limitations

• Primary value is the code implementation rather than a novel dataset
• Metadata does not specify the exact version or subset size of the DUTS dataset used
• Dependent on external DUTS dataset availability for full training

Provenance

Source

CVPR 2019 Paper 'Pyramid Feature Attention Network for Saliency Detection'

Collection Method

Annotated image datasets (DUTS)

Time Range

2019-2024

Freshness

Updated as of July 2024, five years after the original CVPR 2019 publication.