Home News Latest Papers Tutorials Datasets Wiki SOTA LLM Models GPU Leaderboard Events

English

Object Recognition On Shape Bias

Metrics

shape bias

Results

Performance results of various models on this benchmark

Model Name	shape bias	Paper Title	Repository
ViT-22B-384	86.4	Scaling Vision Transformers to 22 Billion Parameters
ResNet-50	22.1	ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
ViT-22B-560	83.8	Scaling Vision Transformers to 22 Billion Parameters
GoogLeNet	31.2	ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
ViT-22B-224	78.0	Scaling Vision Transformers to 22 Billion Parameters
CLIP (ViT-B)	79.9	Learning Transferable Visual Models From Natural Language Supervision
SWSL (ResNet-50)	28.6	Billion-scale semi-supervised learning for image classification
SimCLR (ResNet-50x4)	40.7	A Simple Framework for Contrastive Learning of Visual Representations
Parti	91.7	Intriguing properties of generative classifiers
VGG-16	17.2	ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
Stable Diffusion	92.7	Intriguing properties of generative classifiers
SimCLR (ResNet-50x1)	38.9	A Simple Framework for Contrastive Learning of Visual Representations
SWSL (ResNeXt-101)	49.8	Billion-scale semi-supervised learning for image classification
ResNet-50 (with strong augmentations)	62.2	The Origins and Prevalence of Texture Bias in Convolutional Neural Networks	-
ResNet-50 (L2 eps 5.0 adv trained)	69.5	Do Adversarially Robust ImageNet Models Transfer Better?
SimCLR (ResNet-50x2)	41.7	A Simple Framework for Contrastive Learning of Visual Representations
Imagen	98.7	Intriguing properties of generative classifiers
AlexNet	42.9	ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness

0 of 18 row(s) selected.