bjergshen/mx-maskrcnn

A MXNet implementation of Mask R-CNN

MX Mask R-CNN

A MXNet implementation of Mask R-CNN.

This repository is largely based on the mx-rcnn implementation of Faster RCNN available here.

Main Results

Cityscapes

Method	Training data	Test data	Average	person	rider	car	truck	bus	train	motorcycle	bicycle
Ours	fine-only	test	26.9	33.0	25.7	47.7	21.6	27.4	23.0	19.9	16.9
Reference[5]	fine-only	test	26.2	30.5	23.8	46.9	22.8	32.2	18.6	19.1	16.0
Ours	fine-only	val	31.3	32.6	26.6	49.5	26.5	45.4	32.1	17.6	20.4
Reference[5]	fine-only	val	31.5	-	-	-	-	-	-	-	-

• Backbone: Resnet-50-FPN

COCO

Comming soon, please stay tune.

Requirement

We tested our code on:

Ubuntu 16.04, Python 2.7

numpy(1.12.1), cv2(2.4.9), PIL(4.3), matplotlib(2.1.0), cython(0.26.1), easydict

Preparation for Training

1. Download Cityscapes data (gtFine_trainvaltest.zip, leftImg8bit_trainvaltest.zip). Extract them into 'data/cityscape/'. The folder then look like below:

data/cityscape/
├── leftImg8bit/
│   ├── train/
│   ├── val/
│   └── test/
├── gtFine/
│   ├── train/
│   ├── val/
│   └── test/
└── imglists/
    ├── train.lst
    ├── val.lst
    └── test.lst

2. Download Resnet-50 pretrained model.

bash scripts/download_res50.sh

3. Build MXNet with ROIAlign operator.

cp rcnn/CXX_OP/* incubator-mxnet/src/operator/

To build MXNet from source, please refer to the tutorial.

4. Build related cython code.

make

5. Kick off training

bash scripts/train_alternate.sh

Preparation for Evaluation

1. Prepare Cityscapes evaluation scripts.

bash scripts/download_cityscapescripts.sh

2. Eval

bash scripts/eval.sh

Demo

1. Download model, available at Dropbox/BaiduYun, and place it in model folder.
2. Make sure that you have placed cityscapes data in 'data/cityscapes' folder.

bash scripts/demo.sh

References

1. Tianqi Chen, Mu Li, Yutian Li, Min Lin, Naiyan Wang, Minjie Wang, Tianjun Xiao, Bing Xu, Chiyuan Zhang, and Zheng Zhang. MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems. In Neural Information Processing Systems, Workshop on Machine Learning Systems, 2015
2. Ross Girshick. "Fast R-CNN." In Proceedings of the IEEE International Conference on Computer Vision, 2015.
3. Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. "Faster R-CNN: Towards real-time object detection with region proposal networks." In IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016.
4. Tsung-Yi Lin, Piotr Dollár, Ross Girshick, Kaiming He, Bharath Hariharan, Serge Belongie. "Feature Pyramid Networks for Object Detection." In Computer Vision and Pattern Recognition, IEEE Conference on, 2017.
5. Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick. "Mask R-CNN." In Proceedings of the IEEE International Conference on Computer Vision, 2017.
6. Yangqing Jia, Evan Shelhamer, Jeff Donahue, Sergey Karayev, Jonathan Long, Ross Girshick, Sergio Guadarrama, and Trevor Darrell. "Caffe: Convolutional architecture for fast feature embedding." In Proceedings of the ACM International Conference on Multimedia, 2014.
7. Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. "ImageNet: A large-scale hierarchical image database." In Computer Vision and Pattern Recognition, IEEE Conference on, 2009.
8. Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. "Deep Residual Learning for Image Recognition". In Computer Vision and Pattern Recognition, IEEE Conference on, 2016.
9. Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, Bernt Schiele. "The Cityscapes Dataset for Semantic Urban Scene Understanding." In Computer Vision and Pattern Recognition, IEEE Conference on, 2016.