A carregar...
Publicação
Deep learning tools to study collective behaviour
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 44.8 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
"collectives. The main strength of this approach is the ability to produce very accurate
models. We developed idtracker.ai to extract from a video the trajectory
of each animal in the collective. idtracker.ai is a marker-less multi-animal tracking
system that works by identifying each animal, like its predecessor idTracker.
The difference is that it trains a convolutional neural network in a self-supervised
manner for animal identification. With this strategy, idtracker.ai can track with
high identification accuracy sparse collectives of any species of up to 100 individuals
even if animals touch or occlude each other frequently along the video.
A new tool, idmatcher.ai, works seamlessly with idtracker.ai to identify animals
across different videos. We also tested how deep nets can help to find interaction
rules among animals in collectives.(...)"
Descrição
Palavras-chave
multi-animal tracking system identify animals