Ready usable classifier

[ktugan]

We should upload a readily trained model to a server so easy and fast classifications are possible.

> from bb_tracking import track

> repo = Repository('2015-09-15')
> tracks = track(repo)
# -- model not ready -- #
# -- downloading from http://biorobotics-server.com/bb_tracking_v0.14_model.gzip -- #
> len(tracks)
1337

This would be a nice addition since most people dont need to know what bb_tracking does and only want tracks.

[b2m]

Where

We already have several GitHub Repositories with pretrained models for certain tasks:

• https://github.com/BioroboticsLab/bb_decoder_models
• https://github.com/BioroboticsLab/deeplocalizer_models
• https://github.com/BioroboticsLab/bb_pipeline_models

So I guess we should add another repository with models for the tracking process.

What

At the moment bb_tracking only supports testing features for one stage in a tracking pipeline. Examples for stages in a tracking pipeline could be:

1. Merging Detections to Fragments
2. Anomaly detection for Detections
3. Merging Fragments to bigger Fragments
4. Anomaly detection for Fragments
5. continue with 3 until satisfied

For a practical run on the whole data set we not only need several pre-trained classifiers (one for each type of stage), but also a Scheduler that distributes the data/time ranges to several Walker instances. And afterwards collects the Fragments from each Walker and distributes them in the next step...

My point: it is not only about one model, but about a process that still is in development.