Home
An add-on for DeepLabCut-assisted behavioral analysis.
(Validate model prediction using the GUI. Top: a frame from original video. Bottom: model-prediction of right hindlimb location (y-axis depth) throughout the video. This frame is identified as "slip" by the peak-detection algorithm, based on baseline-corrected model output.)
In Action - Validation, the user can
- import the csv output from a DeepLabCut model (or a spreadsheets containing the same header and column structures as a DLC output file),
- have a frame-by-frame comparison of the model-predicted bodypart location alongside the original video,
- automatically identify "slips" during the ladder rung experiment, including the number of slips, slip depths and the on- and offset of each slip,
- save the predicted slip properties as a csv file,
- manually correct the slip detection results, including removing the false positives and adding undetected slips, and
- save the validated results as a csv file.
We plan to incorporate the main DeepLabCut functions for once the model is trained and deployable. For example, the user would be able to select videos in the GUI for analysis, without having to load the DeepLabCut configurations. The csv output can directly be used for behavioral analyses through the GUI. Manual data validation is possible through a side-by-side comparison of the original video and a graph of the model prediction.
In addition, we hope to add basic group statistics functions, such as ANOVA and t-tests, to further streamline video-based behavioral analysis.