Domain Generalization On Vizwiz

Métriques

Accuracy - All Images

Accuracy - Clean Images

Accuracy - Corrupted Images

Résultats

Résultats de performance de divers modèles sur ce benchmark

Nom du modèle	Accuracy - All Images	Accuracy - Clean Images	Accuracy - Corrupted Images	Paper Title
EfficientNet-B4	41.7	46.4	35.6	EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
ResNet-18 (lpf3)	35.6	39.5	28.5	Making Convolutional Networks Shift-Invariant Again
EfficientNet-B4 (advprop+autoaug)	48.1	51.4	42.5	Adversarial Examples Improve Image Recognition
ResNet-101 (lpf5)	41	45.8	34.8	Making Convolutional Networks Shift-Invariant Again
ResNet-50 (lpf5)	41.5	45.3	35.2	Making Convolutional Networks Shift-Invariant Again
ResNet-101	46.3	50.1	40.5	Deep Residual Learning for Image Recognition
ResNeXt-101 32x16d	51.7	54.8	48.1	Aggregated Residual Transformations for Deep Neural Networks
ResNet-50 (SIN)	25.3	30	20.4	ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
ResNet-50 (augmix)	42.2	46.4	35.9	AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
EfficientNet-B2	38.1	42.8	31.4	EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
VGG-16	34.7	39.5	28.5	Very Deep Convolutional Networks for Large-Scale Image Recognition
ResNet-34 (lpf2)	38.3	42.8	32.4	Making Convolutional Networks Shift-Invariant Again
DenseNet121 (lpf5)	38.7	42.7	32	Making Convolutional Networks Shift-Invariant Again
ResNet-50 (IN-C_brightness)	38.8	43.5	32.5	Measuring Robustness to Natural Distribution Shifts in Image Classification
EfficientNet-B5 (advprop+autoaug)	49.1	51.7	44	Adversarial Examples Improve Image Recognition
ResNet-50 (IN-C_motion_blur)	35.7	39.6	30.2	Measuring Robustness to Natural Distribution Shifts in Image Classification
EfficientNet-B6 (advprop+autoaug)	49.6	53.2	44.7	Adversarial Examples Improve Image Recognition
VGG-11 BN	32.9	37.1	25.8	Very Deep Convolutional Networks for Large-Scale Image Recognition
EfficientNet-B4 (autoaug)	44.3	48.6	38.2	AutoAugment: Learning Augmentation Strategies From Data
VOLO-D5	57.2	59.7	51.8	VOLO: Vision Outlooker for Visual Recognition

0 of 90 row(s) selected.