This project contains programms to tag bees on beesbook images.

Make sure you have OpenCV 2.4, Boost 1.58.0 and Qt5 installed. The code currently depends on Caffe's master branch. Check it out and compile it. To build the code run:

$ mkdir build
$ cd build
$ cmake ..
$ make

The binaries should be now in the build/source/tagger directory.

Built the code and then run in you build directory:

$ ctest

This program lets you create a training dataset.

Create a file with the paths to the images you want to tag. The content of your file should look like this:

Cam_2_20140805143859_2.jpeg
Cam_2_20140805142820_2.jpeg
Cam_2_20140805145011_4.jpeg

In every line stands an image to tag.

Use the find_all_images.sh script to find all images in a directory.

$ ./scripts/find_all_images.sh DIRECTORY > images.txt

Some tags are to near to border of the image. The preprocess utility program adds border and can equalize the histogram with the -use-hist-eq 1 option or create binary images via -binary-image 1.

$ preprocess -o OUTPUT_DIRCTORY images.txt

The new images will be saved to the OUTPUT_DIRECTORY.

The next step is to use the BeesBook pipeline to generate proposals.

Run:

$ generate_proposals FILE_WITH_PATHS

where FILE_WITH_PATHS is the one generated by preprocess generate_proposals creates a .desc file for every image.

Start the actual tagging GUI.

$ tagger FILE_WITH_PATHS

tagger finds the .desc files and updates them as you tag the images.

When you have enough images tagged, you can start to generate a training set:

$ generate_dataset -f hdf5 -o hdf5_output --sample-rate 32 FILE_WITH_PATHS

This will create an hdf5_output directory with maybe multiple .hdf5 files in it.