Blank detection

blank_nonblank model

If you only want to separate blank videos (where no animal is present) from videos that contain an animal, use the blank_nonblank model. Unlike the video species classification models, it does not predict a species; it only outputs the probability that a video is blank.

Architecture

The blank_nonblank model uses the same architecture as the time_distributed model, but there is only one output class as this is a binary classification problem.

Default configuration

The full default configuration is available on Github.

The blank_nonblank model uses the same default configuration as the time_distributed model. For the frame selection, an efficient object detection model called MegadetectorLite is run on all frames to determine which are the most likely to contain an animal. Then the classification model is run on only the 16 frames with the highest predicted probability of detection.

Training data

The blank_nonblank model was trained on all the data used for the time_distributed and european models.

Blank detection with the species models

The time_distributed and slowfast species models also include a blank class, so they can identify blank videos while classifying species. See the time_distributed performance section for an analysis of the blank/non-blank tradeoff.