notesum.ai

Published at December 3

Vision Transformers for Weakly-Supervised Microorganism Enumeration

cs.CV

Released Date: December 3, 2024

Authors: Javier Ureña Santiago¹, Thomas Ströhle², Antonio Rodríguez-Sánchez¹, Ruth Breu²

Aff.: ¹Intelligent and Interactive Systems, University of Innsbruck, Austria; ²Quality Engineering, University of Innsbruck, Austria

Arxiv: http://arxiv.org/pdf/2412.02250v1

Refer to caption

Model name	Variant	Depth	Heads	Dimension	MLP Dim.	Number of parameters ( ${10}^{6}$ )
CNN	Base/Medium/Deep	1 / 2 / 3	-	-	16 / 64 / 256	0.59 / 0.61 / 0.96
ResNet	50/101	16 / 33	-	-	2048	23.53 / 42.54
ViT	Vanilla	12	12	768	3072	87.50
XCiT	S24	24	8	384	1536	49.82
CrossViT	Ti	4	3	96 / 192	384 / 768	3.07
Parallel ViT	Ti	12	3	192	192	5.50
Deep ViT	S	16	12	396	1188	34.91
TransCrowd-G	Vanilla	12	12	768	3072