wuhang / refinenet

RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation

Multipath RefineNet

This is the source code for our paper:

RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation
Guosheng Lin, Anton Milan, Chunhua Shen, Ian Reid
In CVPR 2017

A MATLAB based framework for semantic image segmentation and general dense prediction tasks on images.

Update notes

We did a major update of our code on 23 Dec 2016. If you use an older version, please check out a new copy.

Installation

• Install MatConvNet and CuDNN. We have modified MatConvNet for our task. A modified copy of MatConvNet is provided in ./lib/. You need to compile the provided MatConvNet before running. Details of this modification and compiling can be found in main/my_matconvnet_resnet/README.md.
• An example script for exporting lib paths is
  • main/my_matlab.sh
• Download the following ImageNet pretrained models from MatConvNet and place them in ./model_trained/.
  • imagenet-resnet-50-dag, imagenet-resnet-101-dag, imagenet-resnet-152-dag

Network architecture

You can find the network graphs that illustrate our architecture in the folder net_graphs. Please refer to our paper for more details.

Testing

First download the following trained models and put them in ./model_trained/, then refer to the example scripts for applying the trained model on test images.

• Trained on PASCAL VOC 2012 dataset for object segmentation (ResNet-101):

  • refinenet_res101_voc2012.mat (OneDrive, 426M)

• Trained on Person-Part dataset for object parsing (ResNet-101):

  • refinenet_res101_person_parts.mat (OneDrive, 426M)

• Trained on Cityscapes dataset (ResNet-101) for street scene parsing:

  • refinenet_res101_cityscapes.mat (OneDrive, 400M)

• Trained on SUNRGBD dataset (ResNet-101) for indoor scene understanding:

  • refinenet_res101_sunrgbd.mat (OneDrive, 400M)

• Trained on PASCAL-Context dataset (ResNet-101) for scene understanding:

  • refinenet_res101_pascalcontext.mat (OneDrive, 426M)

• Trained on NYUDv2 dataset (ResNet-101) for indoor scene understanding: (if testing on NYUDv2 images, you may need to remove the white border of the images)

  • refinenet_res101_nyud.mat (OneDrive, 426M)

• Trained on MIT ADE20K dataset (ResNet-152) for scene understanding:

  • refinenet_res152_ade.mat (OneDrive, 456M)

Example scripts for applying these models can be found at: demo_refinenet_test_example_[dataset name].m

• e.g., demo_refinenet_test_example_voc.m, demo_refinenet_test_example_person_parts.m, demo_refinenet_test_example_cityscapes.m

A simplified version (much less configurations) of the test examples for applying the trained models can be found at: demo_test_simple_voc.m and demo_test_simple_city.m.

We also include a demo script to evaluate the trained models, e.g., in terms of IoU scores:

• demo_refinenet_evaluate_voc.m to evaluate the segmentation performance of the trained model;

Training

The following demo is provided for training a RefineNet on your own dataset

• demo_refinenet_train.m

We include the improved version of chained pooling in this code, which may achieve better result than using the above provided models.

Citation

If you find the code useful, please cite our work as

@inproceedings{Lin:2017:RefineNet,
	title = {Refine{N}et: {M}ulti-Path Refinement Networks for High-Resolution Semantic Segmentation},
	shorttitle = {RefineNet: Multi-Path Refinement Networks},
	booktitle = {CVPR},
	author = {Lin, G. and Milan, A. and Shen, C. and Reid, I.},
	month = jul,
	year = {2017}
}

License

For academic usage, the code is released under the permissive BSD license. For any commercial purpose, please contact the authors.