Pytorch code of Sequential/Hierarchical ERFNet with PSPNet for real-time semantic segmentation
Switch branches/tags
Nothing to show
Clone or download
Latest commit 02c87f9 Oct 16, 2018

Sequential/Hierarchical ERF-PSPNet

This code is the Pytorch implementation of Sequential/Hierarchical ERF-PSPNet architectures, which are designed for real-time semantic segmentation.

Hierarchical designs include 4x2 and 3x3 hierarchical archictures. The input resolution is set to 640x480.

The code is tested with Python 3.6, Pytorch 0.4.1, CUDA 8.0. Additional Python packages: numpy, matplotlib, Pillow, torchvision.

For training/deployment, you can also use the environment of [ERFNet] ( or the environment of [PIWISE] (

Network architectures: ERFNet, ERF-PSPNet (ERFNet with PPSNet) Loss functions: Cross entropy, Focal loss Datasets: Cityscapes, Mapillary Vistas Data augmentations: Textual and geometric augmentations


If you use this code in your research, please cite our publications:

Yang, K., Wang, K., Bergasa, L.M., Romera, E., Hu, W., Sun, D., Sun, J., Cheng, R., Chen, T. and López, E., 2018. Unifying Terrain Awareness for the Visually Impaired through Real-Time Semantic Segmentation. Sensors, 18(5), p.1506. (

Yang, K., Bergasa, L.M., Romera, E., Huang, X. and Wang, K., 2018. Predicting polarization beyond semantics for wearable robotics. In IEEE-RAS International Conference on Humanoid Robots (Humanoids2018), Beijing, China, November 2018. (