In this tutorial, we will track a small example dataset, included with anTraX. The dataset consists of a thirty minute recording of a colony of 16 Ooceraea biroi ants, split into 6 video files, each of 5 minutes duration. This dataset is a short segment of the longer J16 benchmark dataset, which is also available for dowload together with all the other benchmark datasets. Unlike these larger datasets, the JS16 dataset is appropriate to track on a laptop/desktop computer in a reasonable time.
All commands in the tutorial are to be entered in the bash terminal of your system, in the same environment anTraX was installed into.
Download the test dataset
git clone http://github.com/Social-Evolution-and-Behavior/anTraX-data.git
Inside the repository there is a directory called ‘JS16’ (the experimental directory). For a full explanation of the structure of the experimental directory, refer to the Preparing data for anTraX section.
Open the antrax app
The dataset includes a pre-configured anTraX session. To explore and change the parameters, open the anTraX configuration app:
antrax configure <path-to-JS16>
For a full explanation of the configuration process and the tracking parameters, refer to the Configuring a tracking session section.
The first step is the tracking. To run it, enter in the terminal:
antrax track <path-to-JS16> --nw 3
--nw 3 option tells anTraX to run 3 parallel tracking threads. For a full listing of the options for the track command, refer to the Runnning the tracking section.
Train a classifier
Once tracking is complete, the next step is to train a blob classifier. The dataset includes an already trained classifier and a set of examples. But for the purpose of this example, let’s run an additional round of training with 3 epochs:
antrax train <path-to-JS16>/antrax_demo/classifier --ne 3
For a full explanation of the training step, including how to generate a training dataset, see the Classifying tracklets section.
Once the classifier is trained, we can classify all the tracklets in the experiment:
antrax classify <path-to-JS16>
For a full listing of the options for the classify command, refer to the Classifying tracklets section.
Run graph propagation
The final step of the algorithm is running the graph propagation, or the ‘solve’ step:
antrax solve <path-to-JS16>
For a full explanation of this step and the all the command options, refer to the Graph propagation section.
Now that tracking is complete, we can verify its accuracy and estimate the tracking error:
antrax validate <path-to-JS16>
See the Validating tracking results section and the anTraX publication for explanation about the validation process.
To debug the tracking, manually fix an important point in the experiment, or just view the tracklet graph, use the graph-explorer app:
antrax graph-explorer <path-to-JS16>
See Using the graph explorer to view and debug ID assignments for details about using the app.