What does PyTorch semantic segmentation open source library semseg look like

Today, I will talk to you about what PyTorch semantic segmentation open source library semseg is like, many people may not know much about it. In order to make you understand better, the editor has summarized the following content for you. I hope you can get something according to this article.

Today, I would like to introduce a new PyTorch-based open source library for semantic segmentation, semseg.

Introduction

Semseg uses PyTorch to implement semantic segmentation / scene parsing open source library. It can help developers to train and test all kinds of semantic segmented data sets.

The library mainly uses ResNet50 / 101Comp152 as the backbone, and it can also be easily changed into other classification network structures.

Networks have been implemented, including PSPNet and PSANet, which ranked first in the 2018 ImageNet scene interpretation Challenge @ ECCV16,LSUN semantic Segmentation Challenge 2017 @ CVPR17 and the WAD driverable region Partition Challenge 2018 @ CVPR18. The sample lab dataset includes the mainstream ADE20K,PASCAL VOC 2012 and Cityscapes.

Ps. The developer of the library is a work of PSPNet and PSANet algorithms.

Bright spot

1. Both multi-thread training and multi-process training are supported, and the latter is very fast (the library attaches more importance to training).

two。 The re-implemented algorithm achieves better results, and the code structure is clear (indicating high code quality).

3. All initialization models, trained models and predicted results can be downloaded (https://drive.google.com/open?id=15wx9vOM0euyizq-M1uINgN0_wjVRf9J3) for developers to directly use or study and compare.

The hardware and software environment recommended by the author:

(4 to 8 graphics cards, it seems that there are no multiple cards, semantic segmentation can not afford to play ~)

Training is simple.

The training of the library is very simple, and only one command is needed after simple configuration.

Sh tool/train.sh ade20k pspnet50

The test is simple

After simply configuring the dataset and model path, only one command is needed:

Sh tool/test.sh ade20k pspnet50

Testing on a single image is also easy, for example:

PYTHONPATH=./ python tool/demo.py-- config=config/ade20k/ade20k_pspnet50.yaml-- image=figure/demo/ADE_val_00001515.jpg TEST.scales'[1.0] 'Performance

The results on the three datasets are as follows:

Note that the time listed by the author was trained on 8 GeForce RTX 2080 Ti.

After reading the above, do you have any further understanding of PyTorch semantic segmentation open source library semseg? If you want to know more knowledge or related content, please follow the industry information channel, thank you for your support.