PyTorch-RPNet

PyTorch implementation of Residual Pyramid Learning for Single-Shot Semantic Segmentation

Network

Segmentation Result

Citing RPNet

Please cite RPNet in your publications if it helps your research:

@ARTICLE{8744483,
author={X. {Chen} and X. {Lou} and L. {Bai} and J. {Han}},
journal={IEEE Transactions on Intelligent Transportation Systems},
title={Residual Pyramid Learning for Single-Shot Semantic Segmentation},
year={2019},
volume={},
number={},
pages={1-11},
keywords={Intelligent vehicles;real-time vision;scene understanding;residual learning.},
doi={10.1109/TITS.2019.2922252},
ISSN={1524-9050},
month={},}

This implementation has been tested on the Cityscapes and CamVid (TBD) datasets. Currently, a pre-trained version of the model trained in CamVid and Cityscapes is available here.

Dataset	Classes ¹	Input resolution	Batch size	Epochs	Mean IoU (%)
CamVid	11	480x360	3	300 x 4	64.82
Cityscapes	19	1024x512	3	300 x 4	70.4 (val)

When referring to the number of classes, the void/unlabeled class is always excluded.
Just for reference since changes in implementation, datasets, and hardware can lead to very different results. Reference hardware: Nvidia GTX 1080ti and an Intel Core i7-7920x 2.9GHz.

The results on Cityscapes dataset

Method	Input Size	Mean IoU	Mean iIoU	fps	FLOPs
ENet	1024*512	58.3	34.4	77	4.03B
ERFNet	1024*512	68.0	40.4	59	25.6B
ESPNet	1024*512	60.3	31.8	139	3.19B
BiSeNet	1036*768	68.4	-	69	26.4B
ICNet	2048*1024	69.5	-	30	-
DeepLab(Mobilenet)	2048*1024	70.71(val)	-	16	21.3B
LRR	2048*1024	69.7	48.0	2	-
RefinNet	2048*1024	73.6	47.2	-	263B
RPNet(ENet)	1024*512	63.37	39.0	88	4.28B
RPNet(ERFNet)	1024*512	67.9	44.9	123	20.7B

Installation

Python 3 and pip.
Set up a virtual environment (optional, but recommended).
Install dependencies using pip: pip install -r requirements.txt.

Examples: Training

sh train.sh

Examples: Testing

python main.py -m test --step 1