Inputs - DeNardoLab/BehaviorDEPOT GitHub Wiki
Data Requirements
BehaviorDEPOT requires a video file and associated pose-estimation file for each session to be run. DeepLabCut is the default pose-estimation software for BehaviorDEPOT. v1.3 includes a conversion script for SLEAP and we are committed to providing support for new pose formats when requested by users.
When performing batch analysis, BehaviorDEPOT accepts a single directory as input. Within the input directory, there should be folders for each session you want to analyze, each containing a video and a pose-estimation file.
Video Format
- Formats currently supported: MP4 and AVI.
*Other formats can be added; please contact authors if you have a particular format you would like to have integrated.
Pose-estimation File
- The default pose-estimation software for BehaviorDEPOT is DeepLabCut (https://github.com/DeepLabCut/DeepLabCut).
- A conversion script allows compatibility with SLEAP (https://sleap.ai)
- Accepted pose-estimation file formats include CSV and H5.
Cue File (Optional)
- Input for cue files is required to be a spreadsheet file (.csv, .xlsx), with each event as a row, the start frame in the first column, and the end frame in the second column.
Converting Human Annotations to BehaviorDEPOT Format (hB)
Input Annotation Format
- BehaviorDEPOT reads in frame-wise annotations of behavior for use in Support Modules.
- The input format should be a table:
- Column 1 contains start frames for each behavior bout.
- Column 2 contains stop frames for each behavior bout.
- Column 3 contains behavior labels (e.g. Freezing, Grooming).
- Multiple behaviors may be contained within the same table.
- Table can be saved as a .MAT or .CSV file.
Converting Annotations
- To convert annotations, open the BehaviorDEPOT GUI.
- Select the 'Convert Human Annotations' button on the right side of the GUI.
- Select the table file (.MAT or .CSV) that you wish to convert.
- Type in the total number of frames in the video.
- An 'hB' file will be saved in the same directory as the input table.
Output Format
- The annotations file will contain a single structure: hBehavior.
- This structure contains bout-wise and vectorized representations of annotated data.
- See Behavior,mat documentation for more details.