Open Access

Table 3

Foundation models used in this study.

Model name Backbone Number of parameters Pretraining dataset EMA
MAE (He et al. 2022) ViT-Base 16×16 86M ImageNet-1k
DINOv2 (Oquab et al. 2024) ViT-Base 14×14 86M LVD-142M (proprietary) +
DINOv1 (Caron et al. 2021) ViT-Base 16×16 86M ImageNet-1k +
MSN (Assran et al. 2022) ViT-Base 16×16 86M ImageNet-1k +
SigLIP (Zhai et al. 2023) ViT-Base 16×16 86M WebLI (English)
AM-RADIO (Ranzinger et al. 2024) ViT-Base 16x16 98.2M various
ResNet 50 (He et al. 2016) ResNet-50 25.6M ImageNet-1k

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.