chainer-fast-rcnn

It aims at reproducing results of fast-rcnn using Chainer. It can be run only with GPU because roi_pooling_2d layer has only GPU implementation.

Requirements

• OpenCV 3.0 with python bindings

• Chainer 1.4

• fast-rcnn

• dlib v18.18

Create symlink

Create a symlink from the location of original fast-rcnn dir to this project's root dir. (The below line assumes a environment variable $FRCN_ROOT has a path to the fast-rcnn source dir.)

$ ln -s $FRCN_ROOT ./

Make sure that all steps written in the Installation (sufficient for the demo) section of README.md in fast-rcnn have been performed.

Convert model

Convert caffemodel to chainermodel.

$ python scripts/load_net.py

Test

First you should prepare a sample image, and then

$ python scripts/forward.py --img_fn sample.jpg --out_fn result.jpg

Result

 'Overstekend wild' St. Janskerkhof Den Bosch © FaceMePLS (https://www.flickr.com/photos/faceme/5891724192)