JAABA is a machine learning-based program that will help you create automatic behavior detectors. To define a behavior and create a behavior detector, you will need to label the behavior of animals in a subset of frames in videos. It is very important to choose these labels carefully. In addition, there are limitations to the types of behaviors for which JAABA can create behavior classifiers. These limitations are not exactly well-defined, and the best way we know for understanding them is to try out JAABA. If you are new to JAABA, please follow the instructions in this document to get a feel for the types of behaviors JAABA can create detectors for, and the effect of different types of labels on JAABA's performance. In general, you will start with a task that we know is within JAABA's capabilities, then iteratively make the task more complicated to test and understand JAABA's abilities and limitations.
sampledata/Chase_JAABAProject.mat and the two experiments
pBDPGAL4U_TrpA_Rig2Plate17BowlA_20120202T145723 and
Chase1_TrpA_Rig1Plate15BowlA_20120404T141155. Read and refer to the rest of the documentation on this site
while training, in particular the sections on Training.html and
Evaluating a Behavior Classifier. While training, we recommend you start with the
most obvious bouts of chases and not chases, then gradually try to get JAABA to work on more difficult examples.We have created a video tutorial to using JAABA that we recommend you watch before trying JAABA yourself. This video shows us creating a chase classifier using the Sample Data distributed on out website. We recommend you watch the video in HD, as details will not be visible at lower resolutions. This sample and data and this tutorial are similar to the introduction we gave in our usability study described in our paper.
We have provided a sample JAABA project and two sample videos, available for download at Sourceforge.
The sample data directory contains two experiment directories that can be used with JAABA:
pBDPGAL4U_TrpA_Rig2Plate17BowlA_20120202T145723 contains video of our control flies. Chase1_TrpA_Rig1Plate15BowlA_20120404T141155 contains video of flies from a GAL4 line
from the dTrpA neural activation screen which chase more than control. Each experiment directory contains the following files:
movie.ufmf: A video of ~10 male and 10 female flies exploring a Fly Bowl. movie.ufmf.ann: Output of Ctrax for the video. registered_trx.mat: Trajectories of the flies in a format compatible with JAABA.perframe: A directory containing the pre-computed per-frame features for JAABA. ctrax_results_movie_*.avi: A short video showing clips of the flies being tracked. Metadata.xml: xml file containing information about how the video was collected.SampleChaseJAABAProject.jab is a sample JAABA project. To test JAABA, we recommend: