Uploaded December 2015 | Updated September 2026, 2 hours ago
Comparison of larval behavioural annotations. Top left panel shows a binary video of a tracking experiment, while the top right panel shows the midline of the larva reconstructed from ECTS. The bottom panel shows the ground truth, JAABA and ESA annotation going from top to bottom. For better visualization all three annotations are shown on the same panel, but height only has significance for the ESA annotation. ESA annotation is probabilistic, height corresponds to the probability that the action is an example of a certain behaviour. For ground truth and JAABA, the annotation is binary (either 1 or 0). The blue vertical line in the middle marks the time corresponding to the video frame in the two top panels. When midline disappears in the top right panel, the vision algorithm could not isolate the larva from the background. For small gaps (less then 15 frames) ECTS was interpolated. Wider gaps could not be reliable interpolated, hence these times were excluded from the ESA analysis.
This research was published in the journal of the Royal Society Interface in the paper: 'Searching for motifs in the behaviour of larval Drosophila melanogaster and Caenorhabditis elegans reveals continuity between behavioural states' by Szigeti et al. The doi link for the article is rsif.royalsocietypublishing.org/lookup/doi/10.1098/rsif.2015.0899
Comparison of larval behavioural annotations. Top left panel shows a binary video of a tracking experiment, while the top right panel shows the midline of the larva reconstructed from ECTS. The bottom panel shows the ground truth, JAABA and ESA annotation going from top to bottom. For better visualization all three annotations are shown on the same panel, but height only has significance for the ESA annotation. ESA annotation is probabilistic, height corresponds to the probability that the action is an example of a certain behaviour. For ground truth and JAABA, the annotation is binary (either 1 or 0). The blue vertical line in the middle marks the time corresponding to the video frame in the two top panels. When midline disappears in the top right panel, the vision algorithm could not isolate the larva from the background. For small gaps (less then 15 frames) ECTS was interpolated. Wider gaps could not be reliable interpolated, hence these times were excluded from the ESA analysis.
This research was published in the journal of the Royal Society Interface in the paper: 'Searching for motifs in the behaviour of larval Drosophila melanogaster and Caenorhabditis elegans reveals continuity between behavioural states' by Szigeti et al. The doi link for the article is rsif.royalsocietypublishing.org/lookup/doi/10.1098/rsif.2015.0899
