Fine Grained Image Classification On Oxford

Metrics

Accuracy

FLOPS

PARAMS

Results

Performance results of various models on this benchmark

Model Name	Accuracy	FLOPS	PARAMS	Paper Title	Repository
ResNet50 (A1)	97.9%	4.1	24M	ResNet strikes back: An improved training procedure in timm	-
Grafit (RegNet-8GF)	99.1%	-	-	Grafit: Learning fine-grained image representations with coarse labels	-
AutoFormer-S \| 384	-	-	-	AutoFormer: Searching Transformers for Visual Recognition	-
Wide-ResNet-101 (Spinal FC)	99.30%	-	-	SpinalNet: Deep Neural Network with Gradual Input	-
DenseNet-201	98.29	-	-	A Comprehensive Study on Torchvision Pre-trained Models for Fine-grained Inter-species Classification	-
PC Bilinear CNN	93.65%	-	-	Pairwise Confusion for Fine-Grained Visual Classification	-
ResMLP-12	97.4%	-	-	ResMLP: Feedforward networks for image classification with data-efficient training	-
NAT-M3	98.1	250M	3.7M	Neural Architecture Transfer	-
µ2Net (ViT-L/16)	99.61%	-	-	An Evolutionary Approach to Dynamic Introduction of Tasks in Large-scale Multitask Learning Systems	-
TNT-B	99.0%	-	65.6M	Transformer in Transformer	-
DenseNet-201(Spinal FC)	98.36	-	-	A Comprehensive Study on Torchvision Pre-trained Models for Fine-grained Inter-species Classification	-
NAT-M1	-	152M	3.3M	Neural Architecture Transfer	-
BiT-M (ResNet)	99.30%	-	-	Big Transfer (BiT): General Visual Representation Learning	-
ResMLP-24	97.9%	-	-	ResMLP: Feedforward networks for image classification with data-efficient training	-
CCT-14/7x2	-	15G	22.5M	Escaping the Big Data Paradigm with Compact Transformers	-
AutoAugment	95.36%	-	-	AutoAugment: Learning Augmentation Policies from Data	-
IELT	99.64%	-	-	Fine-Grained Visual Classification via Internal Ensemble Learning Transformer
NAT-M2	97.9	195M	3.4M	Neural Architecture Transfer	-
FixInceptionResNet-V2	95.7%	-	-	Fixing the train-test resolution discrepancy	-
Assemble-ResNet	98.9%	-	-	Compounding the Performance Improvements of Assembled Techniques in a Convolutional Neural Network	-

0 of 25 row(s) selected.