DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation

TLDR

The paper proposes DFANet, an extremely efficient CNN architecture for real‑time semantic segmentation under resource constraints. DFANet uses a single lightweight backbone that aggregates discriminative features via sub‑network and sub‑stage cascades, employing multi‑scale feature propagation to reduce parameters while preserving receptive field and learning capacity, thereby balancing speed and accuracy. Experiments on Cityscapes and CamVid demonstrate that DFANet achieves 70.3 % mIOU on Cityscapes test with only 1.7 GFLOPs and 160 FPS, and 71.3 % mIOU with 3.4 GFLOPs on higher resolution, outperforming state‑of‑the‑art real‑time methods with 8× fewer FLOPs and twice the speed while maintaining comparable accuracy.

Abstract

This paper introduces an extremely efficient CNN architecture named DFANet for semantic segmentation under resource constraints. Our proposed network starts from a single lightweight backbone and aggregates discriminative features through sub-network and sub-stage cascade respectively. Based on the multi-scale feature propagation, DFANet substantially reduces the number of parameters, but still obtains sufficient receptive field and enhances the model learning ability, which strikes a balance between the speed and segmentation performance. Experiments on Cityscapes and CamVid datasets demonstrate the superior performance of DFANet with 8$\times$ less FLOPs and 2$\times$ faster than the existing state-of-the-art real-time semantic segmentation methods while providing comparable accuracy. Specifically, it achieves 70.3\% Mean IOU on the Cityscapes test dataset with only 1.7 GFLOPs and a speed of 160 FPS on one NVIDIA Titan X card, and 71.3\% Mean IOU with 3.4 GFLOPs while inferring on a higher resolution image.