Awesome

Recurrent Scale Approximation (RSA) for Object Detection

Codebase for Recurrent Scale Approximation for Object Detection in CNN published at ICCV 2017, [arXiv]. Here we offer the training and test code for two modules in the paper, scale-forecast network and recurrent scale approximation (RSA). Models for face detection trained on some open datasets are also provided.

<span style="color:red">Note: This project is still underway. Please stay tuned for more features soon!</span>

Codebase at a Glance

train/: Training code for modules scale-forecast network and RSA

predict/: Test code for the whole detection pipeline

afw_gtmiss.mat: Revised face data annotation mentioned in Section 4.1 in the paper.

Grab and Go (Demo)

Caffe models for face detection trained on popular datasets.

Base RPN model: predict/output/ResNet_3b_s16/tot_wometa_1epoch, trained on Widerface (fg/bg), COCO (bg only) and ImageNet Det (bg only)
RSA model: predict/output/ResNet_3b_s16_fm2fm_pool2_deep/65w, trained on Widerface, COCO, and ImageNet Det

Steps to run the test code:

Compile CaffeMex_v2 with matlab interface
Add CaffeMex_v2/matlab/ to matlab search path
See tips in predict/script_start.m and run it!
After processing for a few minutes, the detection and alignment results will be shown in an image window. Please click the image window to view all results. If you set line 8 in script_start.m to false as default, you should observe some results as above.

Train Your Own Model

Still in progress, this part will be released later.

FAQ

We will list the common issues of this project as time goes. Stay tuned! :)

Citation

Please kindly cite our work if it helps your research:

@inproceedings{liu_2017_rsa,
  Author = {Yu Liu and Hongyang Li and Junjie Yan and Fangyin Wei and Xiaogang Wang and Xiaoou Tang},
  Title = {Recurrent Scale Approximation for Object Detection in CNN},
  Journal = {IEEE International Conference on Computer Vision},
  Year = {2017}
}

Acknowledgment

We appreciate the contribution of the following researchers:

Dong Chen @Microsoft Research, some basic ideas are inspired by him when Yu Liu worked as an intern at MSR.

Jiongchao Jin @Beihang University, some baseline results are provided by him.