[jsk_perception] Scripts for Bag of Features Object Recognition

[wkentaro]

These scripts are moved from start-jsk/2014-semi and rewrited to be used in more general cases.

[wkentaro]

Tutorial

Download raw data

$ roscd jsk_data
$ make large KEYWORD:=jsk_2014_picking_challenge/20150428_collected_images

Create sift dataset

$ roslaunch jsk_perception create_feature0d_dataset.launch container_path:=`rospack find jsk_data`/large/20150428_collected_images

Extract bag of features & its histogram

$ rosrun jsk_perception extract_bof.py fit `rospack find jsk_perception`/trained_data/feature0d_dataset.pkl.gz -O `rospack find jsk_perception`/trained_data/bof.pkl.gz
$ rosrun jsk_perception extract_bof.py dataset `rospack find jsk_perception`/trained_data/feature0d_dataset.pkl.gz `rospack find jsk_perception`/trained_data/bof.pkl.gz -O `rospack find jsk_perception`/trained_data/bof_hist.pkl.gz

Train classifier

$ rosrun jsk_perception simple_classifier_trainer.py `rospack find jsk_perception`/trained_data/bof_hist.pkl.gz -O `rospack find jsk_perception`/trained_data/clf.pkl.gz

Run for real objects

$ wget https://i.warosu.org/data/ck/img/0048/57/1381305564234.jpg -O /tmp/oreo_mega_stuf_sample.jpg
$ roscore
$ rosrun jsk_perception image_publisher.py _file_name:=/tmp/oreo_mega_stuf_sample.jpg
$ rosrun imagesift imagesift image:=/image_publisher/output
$ rosrun jsk_perception extract_bof.py extract `rospack find jsk_perception`/trained_data/bof.pkl.gz
$ rosrun jsk_perception simple_classifier.py _clf_path:=`rospack find jsk_perception`/trained_data/clf.pkl.gz

$ rostopic echo /simple_classifier/output
data: oreo_mega_stuf
---
data: oreo_mega_stuf
---
...

[garaemon]

Write down it in README.md

[wkentaro]

I added description to readme like below:

---