Self-supervised learning on MeerKAT wide-field continuum images

Open Access

Table 3

Backbones referenced in this work.

Backbone name	Source	Training dataset	Training	Training	Backbone
			samples	method	network
MGCLS Resnet	This work	MGCLS_20k	20 K	DINO	Resnet50
MGCLS ViTS	This work	MGCLS_5k	5K	DINO	ViT-S8
RGZ BYOL	Slijepcevic et al. (2024)	Radio Galaxy Zoo	108 K	BYOL	Resnet18
GZ2 MoCo	Hayat et al. (2021)	Galaxy Zoo 2	1.2M	MoCo v2	Resnet50
DINO Resnet	Caron et al. (2021)	ImageNet	1.3 M	DINO	Resnet50
DINOv2 ViTG	Oquab et al. (2024)	DINOv2	143 M	DINOv2	ViT-G14
DeCALS BYOL	Mohale & Lochner (2024)	Galaxy Zoo DeCALS	270 K	BYOL	Resnet18
RGZ VDVAE	Andrianomena & Tang (2024)	Radio Galaxy Zoo	108 K	VDVAE	Resnet34

Notes. All of these backbones are trained in a self-supervised manner, even if they are trained on different datasets and with different network architectures.

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.