Skip to content

Latest commit

 

History

History
64 lines (49 loc) · 5.94 KB

README.md

File metadata and controls

64 lines (49 loc) · 5.94 KB

[WACV2023] Dense Prediction with Attentive Feature Aggregation

This is the official implementation of our paper "Dense Prediction with Attentive Feature Aggregation".

Yung-Hsu Yang, Thomas E. Huang, Min Sun, Samuel Rota Bulò, Peter Kontschieder, Fisher Yu

[Paper] [Project]

Abstract

Aggregating information from features across different layers is essential for dense prediction models. Despite its limited expressiveness, vanilla feature concatenation dominates the choice of aggregation operations. In this paper, we introduce Attentive Feature Aggregation (AFA) to fuse different network layers with more expressive non-linear operations. AFA exploits both spatial and channel attention to compute weighted averages of the layer activations. Inspired by neural volume rendering, we further extend AFA with Scale-Space Rendering (SSR) to perform a late fusion of multi-scale predictions. AFA is applicable to a wide range of existing network designs. Our experiments show consistent and significant improvements on challenging semantic segmentation benchmarks, including Cityscapes and BDD100K at negligible computational and parameter overhead. In particular, AFA improves the performance of the Deep Layer Aggregation (DLA) model by nearly 6% mIoU on Cityscapes. Our experimental analyses show that AFA learns to progressively refine segmentation maps and improve boundary details, leading to new state-of-the-art results on boundary detection benchmarks on NYUDv2 and BSDS500.

Installation

Please refer to INSTALL.md for installation and to PREPARE_DATASETS.md for dataset preparation.

Get Started

Please see GETTING_STARTED.md for the basic usage.

Model Zoo

Cityscapes

Model Crop Size Batch Size Training Epochs mIoU (val) mIoU (test) config weights Preds Visuals
AFA-DLA (Train) 1024x2048 8 375 85.14 - config model val val
AFA-DLA (Train + Val) 1024x1024 16 275 - 83.58 config model test test

BDD100K

Model Crop Size Batch Size Training Epochs mIoU (val) mIoU (test) config weights Preds Visuals
AFA-DLA 720x1280 16 200 67.46 58.70 config model val | test val | test

NYUDv2

Model Crop Size Batch Size Training Epochs ODS OIS config weights Preds Visuals
AFA-DLA (RGB) 480x480 16 54 0.762 0.775 config model test test
AFA-DLA (HHA) 480x480 16 54 0.718 0.730 config model test test

BSDS500

Model Crop Size Batch Size Training Epochs ODS OIS config weights Preds Visuals
AFA-DLA 416x416 16 14 0.812 0.826 config model test test
AFA-DLA (PASCAL) 416x416 16 20 0.810 0.826 config model test test

Qualitative Results

Cityscapes Test Set

BDD100K Test Set

You can find more visualizations in our project page.

Citation

@inproceedings{yang2023dense,
    title={Dense prediction with attentive feature aggregation},
    author={Yang, Yung-Hsu and Huang, Thomas E and Sun, Min and Bul{\`o}, Samuel Rota and Kontschieder, Peter and Yu, Fisher},
    booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
    pages={97--106},
    year={2023}
}

Acknowledgement

The codbase is developed from NVIDIA segmentation. We deeply thank for the help of their open-sourced code.