RSA pycaffe

pycaffe port of RSA(Recurrent Scale Approximation for Object Detection)

Updata 2017/12/12

aligned images with (ec-mc-y= 40,size=[96,112]) using MTCNN & RSA

MTCNN:

RSA:

LFW landmark Points using RSA:

here

Usage

1.download matlab version of RSA

2.copy RSA.py,nms.py to ./predict/

3.make pycaffe under your_caffe_path

4.change the caffe_root in RSA.py to your_caffe_path/python

5.create a sample of RSA,e.g,rsa = RSA()

6.call function detect

input param:image
output param:score,box,pts

e.g

img = cv2.imread(img_path,1)
score,box,pts = rsa.detect(image)

update 2017/12/9:

7.For multiGPUs:

1)Change work function in Work.py to what you want

2)run. e.g:

import Multi_GPUs
list = ['1.jpg','2.jpg','3.jpg']
gpu_ids = [0,1,2]
multiprocess(list,gpu_ids)

Attention

There will be some different while using imread between opencv and matlab.The pixel value will be different if you use opencv version hight than opencv2.4.

Thanks

my friend and pre-fellow @Dickachu Yang helps me a lot on python numpy module

His pycaffe port of RSA:

https://github.com/power0341/RSA_pycaffe

Ps:

my first python code,very ugly.

A lot of thing to learn with python and numpy ^_^