JOSS: Automated detection of particles

[mhubii]

Sorry as somewhat of a newbie here..

I ran Detection -> Use Example Model -> Detect with default settings. (https://particletracking.readthedocs.io/en/latest/RodTracker/RodTracker.html#automated-detection-of-particles)

The result only detected sporadically, see below

rodtracker_results.mp4

Any explanation for why the presented method does not auto-detect, as claimed? I assume with depth information and instance segmentation, this should be possible. Please advise.

Refers to openjournals/joss-reviews#5986

[dmitrypuzyrev]

Hello @mhubii,
can you please describe what steps you've made here exactly?
It seems that you've run 2D detection for one camera view for blue color, and then you are plotting only rod number 0.
In each frame on your video this corresponds to one of the detected objects, but it jumps from object to object because detected rods are not yet sorted (tracked).
Because tracking/view matching is performed in 3D, to get reasonable tracking results you also need to run detection for the second view, load the example stereo calibration data and then run detection on selected frame:
https://particletracking.readthedocs.io/en/latest/RodTracker/RodTracker.html#rod-tracking-and-3d-coordinate-reconstruction
Did you perform those steps as well?

[mhubii]

no worries, so I just followed the instructions and left configurations untouched.

On data: RodTracker/src/RodTracker/resources/example_data/images/gp3

[dmitrypuzyrev]

So if you indeed used only one camera data, this behavior is expected, because no 2D tracking was performed at all.
2D tracking would be indeed possible too, but then there would be problems with tracking rods during overlapping and complications with 3D matching too.

By the way, to test the 3D tracking, please use the latest updates on develop branch, there was an issue that prevented proper tracking (Issue #96) which was recently fixed. We will merge the fix to the other branches as well soon.

[mhubii]

I mean the tutorial says "Automated detection of particle", so tracking is the expected outcome.

Sorry, but I won't use the develop branch. As per JOSS, this package is expected to work out of the box.

[dmitrypuzyrev]

Hi @mhubii,
now the default number of detected particles is increased to 25 for each color (according to the experiment shown on example images).
Most of the particles are automatically detected on the frames.
Please check that it works as expected for you!

[mhubii]

I'll check tomorrow. Sorry for the delay

[mhubii]

this seems fixed