Res2Net: a new multi-scale backbone architecture

Encadrants : Vannary Meas-Yedid
Disponible : NON
Nombre d'étudiants : 1
Description :

The human vision is generally performed at multi-scale levels. That is why representing features at multiple scales is of great importance for many vision tasks. Recent advances in backbone convolutional neural networks (CNN) continually demonstrate stronger multi-scale representation ability, leading to consistent performance gains on a wide range of applications. However, most existing methods represent the multi-scale features in a layer-wise manner.

Travail demandé : In this project, after a comprehensive review of the literature related multi-scale representation, the student should implement the Res2Net architecture proposed by [1] and test it on biological images for many tasks such as object detection and instance segmentation.
Liens complémentaires : Shang-Hua Gao, Ming-Ming Cheng, Kai Zhao, Xin-Yu Zhang, Ming-Hsuan Yang, Philip Torr
Res2Net: a new multi-scale backbone architecture. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, no. 2, pp. 652–662, 2021