delldu/maskrcnn

Mask R-CNN is for "instance segmentation". Please reference https://arxiv.org/abs/1703.06870.

MaskRCNN

Mask R-CNN is for "instance segmentation". Please reference https://arxiv.org/abs/1703.06870.

Example 1

Example2

Demo

python predict.py images/car58a54312d.jpg

Training

1. Put Coco files under data directory.

data/
├── annotations
├── test2014
├── train2014
└── val2014

2. ./train.sh

Evaluation

./eval.sh

DONE (t=2.57s).
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.317  
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.525  
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.336  
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.139  
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.366  
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.492  
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.261  
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.369  
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.379  
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.169  
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.425  
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.562  
Prediction time: 349.81333112716675. Average 0.6996266622543335/image
Total time:  401.12283730506897

Requirements

• Python 3.6.2

• Pytorch 1.0.0

• matplotlib, scipy, scikit-image

pip install scipy==1.2.1

Installation

1. Clone this repository.

    git clone https://github.com/delldu/MaskRCNN.git
   

2. Download pre-trained model.

　　　Download mask_rcnn_coco.pth from https://pan.baidu.com/s/1HVUdfrFKPMGlMcUP7mXZGw

and put it under models .

3. Install c++ extension packages

   cd c++ext  
   make  
   cd ../
   
   cd cocoapi/PythonAPI  
   make  
   cd ../..
   

Thanks

1. Mask R-CNN https://arxiv.org/abs/1703.06870

2. https://github.com/multimodallearning/pytorch-mask-rcnn

Chinese Document

中文文档

maskrcnn's Issues

Issue with evaluation mode Precision and recall

while in debugging mode, in linear layer of the model gives the class_id all zeros. input size is (91, 1024), weight is (81,1024) and bias is 81. when doing the torch.addmm(bias, input, weight.t()) is gives the all class_id 0 0 0 0 0 0 00 ......... i spend about a week on this issues. please help.

How to train my own datasets? Thanks.

How train VOC2007?