No X/Y Coordinate Feature Extraction from DLC: Only Likelihood

[jeburling]

I am in the process of creating a Simba project to identify freezing. I get through the entire configuration file creation, video parameter specification, and outlier correction. However, when I get to the extract features section, simba only extracts the likelihood from my DLC csv data files, not the x/y spatial measures.

When I go into my 'input_csv' folder in the simba project folder, the file contains x/y spatial measures and the likelihood values (as the DLC project normally generates).

However, when I extract features, the csv files only contain the likelihood measures (see below)

Is there a way to fix this? My simba project file utilizes classical tracking with a user_defined body part configuration (copied from my DLC). Happy to provide any further details.

[sronilsson]

Hey @jeburling - interesting thanks for reporting this and let's figure it out.

To understand where in the pipeline the values become zero:

Do you see the 0s inside the project_folder\csv\outlier_corrected_movement_location and project_folder\csv\outlier_corrected_movement directories too? Could you send me a screengrab of the project_folder\csv\outlier_corrected_movement_location file too?

EDIT:

Are the body-part coordinates zero in the entire files in the feature extraction files? I see the confidence value are very low in the first 30ish frames just to be sure it isn't something to do with that and they are filtered out (although I cant immediately see how that would come about).

[jeburling]

The 0s are in the project_folder\csv\outlier_corrected_movement_location (see below)

However, they do not appear in the project_folder\csv\outlier_corrected_movement (see below)

[sronilsson]

Thank you @jeburling - could be a bug I've introduced in outlier correctiopn, will check - one more question: which version of simba are you running and is shown when typing pip show simba-uw-tf-dev ?

To confirm, if you click to skip outlier correction and then click again to extract features, do the values show up?

[sronilsson]

PS. Also try to increase the outlier location criterion value. I think what could be happening is that the first frame is detected as an outlier. As there is no prior value (as the first frame is an outlier) to reset the body-parts to, SimBA is confused and sets it to a 0.

[jeburling]

When I skipped outlier correction it worked! I will go back and try outlier correction with the increased outlier location criterion value if I decide to do an outlier correction. Thank you so much for your help!

[sronilsson]

Thanks for letting me know @jeburling - let me know if anything else comes up.

... just beware of the low detection confidences. This suggest that the animal isn't appropriately detected by the pose-estimation model, either because the animal isn't there yet in the beginning of the video, or the model is not very good. Because the animal can't be found in the video frames, the behavior model may not be very good.