Early prediction of the transferability of bovine embryos from videomicroscopy
Résumé
We aim to predict early and automatically the transferability of embryos for biological issues and cattle breeding challenges, taking as input 2D time-lapse videos up to the fourth embryonic development day. We propose: 1- a formulation as a supervised classification with two classes Transferable (T) and Non Transferable (NT), 2- a 3D-CNN with three pathways, multi-scale in time and able to handle appearance and motion in different ways, 3- the use of the focal loss for training. Our model SFR compares favorably to other methods and demonstrates its accuracy and efficiency for our challenging biological task.
Origine | Fichiers produits par l'(les) auteur(s) |
---|