Semantic Segmentation on MIT ADE20K dataset in PyTorch

This is a PyTorch implementation of semantic segmentation models on MIT ADE20K scene parsing dataset (http://sceneparsing.csail.mit.edu/).

ADE20K is the largest open source dataset for semantic segmentation and scene parsing, released by MIT Computer Vision team. Follow the link below to find the repository for our dataset and implementations on Caffe and Torch7: https://github.com/CSAILVision/sceneparsing

If you simply want to play with our demo, please try this link: http://scenesegmentation.csail.mit.edu You can upload your own photo and parse it!

You can also use this colab notebook playground here to tinker with the code for segmenting an image.

All pretrained models can be found at: http://sceneparsing.csail.mit.edu/model/pytorch

[From left to right: Test Image, Ground Truth, Predicted Result]

Color encoding of semantic categories can be found here: https://docs.google.com/spreadsheets/d/1se8YEtb2detS7OuPE86fXGyD269pMycAWe2mtKUj2W8/edit?usp=sharing

Highlights

Syncronized Batch Normalization on PyTorch

This module computes the mean and standard-deviation across all devices during training. We empirically find that a reasonable large batch size is important for segmentation. We thank Jiayuan Mao for his kind contributions, please refer to Synchronized-BatchNorm-PyTorch for details.

The implementation is easy to use as:

• It is pure-python, no C++ extra extension libs.
• It is completely compatible with PyTorch's implementation. Specifically, it uses unbiased variance to update the moving average, and use sqrt(max(var, eps)) instead of sqrt(var + eps).
• It is efficient, only 20% to 30% slower than UnsyncBN.

Dynamic scales of input for training with multiple GPUs

For the task of semantic segmentation, it is good to keep aspect ratio of images during training. So we re-implement the DataParallel module, and make it support distributing data to multiple GPUs in python dict, so that each gpu can process images of different sizes. At the same time, the dataloader also operates differently.

Now the batch size of a dataloader always equals to the number of GPUs, each element will be sent to a GPU. It is also compatible with multi-processing. Note that the file index for the multi-processing dataloader is stored on the master process, which is in contradict to our goal that each worker maintains its own file list. So we use a trick that although the master process still gives dataloader an index for __getitem__ function, we just ignore such request and send a random batch dict. Also, the multiple workers forked by the dataloader all have the same seed, you will find that multiple workers will yield exactly the same data, if we use the above-mentioned trick directly. Therefore, we add one line of code which sets the defaut seed for numpy.random before activating multiple worker in dataloader.

State-of-the-Art models

• PSPNet is scene parsing network that aggregates global representation with Pyramid Pooling Module (PPM). It is the winner model of ILSVRC'16 MIT Scene Parsing Challenge. Please refer to https://arxiv.org/abs/1612.01105 for details.
• UPerNet is a model based on Feature Pyramid Network (FPN) and Pyramid Pooling Module (PPM). It doesn't need dilated convolution, an operator that is time-and-memory consuming. Without bells and whistles, it is comparable or even better compared with PSPNet, while requiring much shorter training time and less GPU memory. Please refer to https://arxiv.org/abs/1807.10221 for details.
• HRNet is a recently proposed model that retains high resolution representations throughout the model, without the traditional bottleneck design. It achieves the SOTA performance on a series of pixel labeling tasks. Please refer to https://arxiv.org/abs/1904.04514 for details.

Reference

If you find the code or pre-trained models useful, please cite the following papers:

Semantic Understanding of Scenes through ADE20K Dataset. B. Zhou, H. Zhao, X. Puig, T. Xiao, S. Fidler, A. Barriuso and A. Torralba. International Journal on Computer Vision (IJCV), 2018. (https://arxiv.org/pdf/1608.05442.pdf)

@article{zhou2018semantic,
  title={Semantic understanding of scenes through the ade20k dataset},
  author={Zhou, Bolei and Zhao, Hang and Puig, Xavier and Xiao, Tete and Fidler, Sanja and Barriuso, Adela and Torralba, Antonio},
  journal={International Journal on Computer Vision},
  year={2018}
}

Scene Parsing through ADE20K Dataset. B. Zhou, H. Zhao, X. Puig, S. Fidler, A. Barriuso and A. Torralba. Computer Vision and Pattern Recognition (CVPR), 2017. (http://people.csail.mit.edu/bzhou/publication/scene-parse-camera-ready.pdf)

@inproceedings{zhou2017scene,
    title={Scene Parsing through ADE20K Dataset},
    author={Zhou, Bolei and Zhao, Hang and Puig, Xavier and Fidler, Sanja and Barriuso, Adela and Torralba, Antonio},
    booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
    year={2017}
}