Skip to content

Latest commit

 

History

History
43 lines (22 loc) · 3.8 KB

validation_tutorial.md

File metadata and controls

43 lines (22 loc) · 3.8 KB
Raw

Validate model on single video

  1. Click on File --> Load project, and the following window will pop up.

  1. Click Browse File and select the project_config.ini file and click Load Project.

  2. Under [Run machine model] tab --> validate Model on Single Video, select your features file (.csv). It should be located in project_folder/csv/features_extracted.

Note: If you check the 'Generate Gantt plot' box, the final validation will include two seperate videos - 1 of the behavior and 1 of the gantt plot. However, this takes longer and we suggest leaving it unchecked unless you really want the gantt plot. This may or may not be combined to a single video in the future!

  1. Select a model file (.sav).

  2. Click on Run Model.

  3. Once, it is completed, it should print "Predictions generated.", now you can click on Generate plot. A graph window and a frame window will pop up.

  • Graph window: model prediction probability versus frame numbers will be plot. The graph is interactive, click on the graph and the frame window will display the selected frames.

  • Frame window: Frames of the chosen video with controls.

  1. Click on the points on the graph and picture displayed on the other window will jump to the corresponding frame. There will be a red line to show the points that you have clicked.

  1. Once it jumps to the desired frame, you can navigate through the frames to determine if the behavior is present. This step is to find the optimal threshold to validate your model.

  1. Once the threshold is determined, enter the threshold into the Discrimination threshold entry box and the desire minimum behavior bouth length into the Minimum behavior bout lenght(ms) entrybox.

  • Discrimination threshold: The level of probability required to define that the frame belongs to the target class. Accepts a float value between 0.0-1.0. For example, if set to 0.50, then all frames with a probability of containing the behavior of 0.5 or above will be classified as containing the behavior. For more information on classification theshold, click here

  • Minimum behavior bout length (ms): The minimum length of a classified behavioral bout. Example: The random forest makes the following attack predictions for 9 consecutive frames in a 50 fps video: 1,1,1,1,0,1,1,1,1. This would mean, if we don't have a minimum bout length, that the animals fought for 80ms (4 frames), took a brake for 20ms (1 frame), then fought again for another 80ms (4 frames). You may want to classify this as a single 180ms attack bout rather than two separate 80ms attack bouts. With this setting you can do this. If the minimum behavior bout length is set to 20, any interruption in the behavior that is 20ms or shorter will be removed and the behavioral sequence above will be re-classified as: 1,1,1,1,1,1,1,1,1 - and instead classified as a single 180ms attack bout.

  • Create Gantt plot: Determins if a Gantt chart displaying the detected behavioral bouts should be appended to the validation video. Select None and no Gantt chart will be created (fastest option). Select Gantt chart: video and a temporally aligned Gantt chart video will be appended to your output video (slowest option). Select Gantt chart: final frame only (slightly faster) and a single Gantt image representing all detected behavior bouts in the video will be appended to the output video.

  1. Click Validate to validate your model. You can follow the progress in the main SimBA terminal window. The validation video will be saved in the project_folder/frames/output/validation directory.