2D Backbones

Models

AlexNet

VGG

ResNet

ConvNeXt

EfficientNet

Experiments' configurations

Dataset: CUB_200_2011(birds classification).

Inference timing: end-to-end, 2500 iterations with an input being a 1500x1500x3 uint8 tensor.

Training's configurations :

	alexnet	vgg 11	resnet 18	convnext base	efficientnet b0
batch size	512	256	512	64	256
learning rate	6e-4	7e-4	1e-3	6e-4	2e-3
epochs	30	30	30	30	30

Results

Train and Test logs

alexnet.out

vgg.out

resnet.out

convnext.out

efficientnet.out

Accuracy vs Performance on CPU vs Model size

CPU : Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz

Accuracy vs Performance on GPU vs Model size

GPU: NVIDIA Quadro RTX 6000

Attribution

Most of the code here was inspired by bentrevett/pytorch-image-classification. As such, the LICENSE found there is also included here.