Video Recording
The higher the quality of your video recordings, the better the pose estimation and behavioral classification will be. That said, video recordings where the pups are visible for part of the video and the resolution is high enough to visually identify pups should be good enough. On the other hand, this analysis pipeline can't perform magic, so it will not be able to track dams or pups (or extract behavior) when they are occluded. Some tips for setting up videos to give yourself the best chance of getting high-quality videos for use in AMBER are below.
-->When possible, video recordings should be from side-view from the long end of the cage closest to the nest.
-->Set up the cameras so that the majority of the frame is of the inside of the cage rather than the area outside of the cage.
-->SimBA uses the width of the cage top about half way back to calculate pixels per mm. This step allows for normalization due to different camera distance form the cage or different frame resolutions. While you may be able to normalize to a different known distance, we recommend trying to include this distance in view in your recordings if possible. If this is not possible, we recommend normalizing to a distance that is about at the halfway point in terms of depth of the cage in order to be as similar as possible to the videos used to train the models.
-->SimBA models were trained on videos with a resolution of 920 X 550 at 30 fps, but other resolutions and frame rates are likely to work.
-->All videos used to train the pose estimation models were in greyscale, so the models will not work well on videos recorded in color. Color videos can be converted to greyscale prior to analysis.
Preprocessing videos prior to analysis can be used to change recording resolution, crop videos, down sample frame rate, and convert videos to grey scale after recording. SimBA provides an easy-to-use interface for video preprocessing.