Benchmark and model zoo¶
Common settings¶
All models were trained on
CUB_200_2011_train
and tested on theCUB_200_2011_test
.For fair comparison with other codebases, we report the GPU memory as the maximum value of
torch.cuda.max_memory_allocated()
for all 8 GPUs. Note that this value is usually less than whatnvidia-smi
shows.All pytorch-style pretrained backbone are form PyTorch model zoo.
Backbone models¶
The detailed table of the commonly used backbone models in FGVCLib is listed below:
model | source | link | description |
---|---|---|---|
ResNet18 | TorchVision | torchvision's ResNet-18 | From torchvision's ResNet-18. |
ResNet34 | TorchVision | torchvision's ResNet-34 | From torchvision's ResNet-34. |
ResNet50 | TorchVision | torchvision's ResNet-50 | From torchvision's ResNet-50. |
ResNet101 | TorchVision | torchvision's ResNet-101 | From torchvision's ResNet-101. |
ResNet152 | TorchVision | torchvision's ResNet-152 | From torchvision's ResNet-152. |
Vgg11 | TorchVision | torchvision's Vgg-11 | From torchvision's Vgg-11. |
Vgg13 | TorchVision | torchvision's Vgg-13 | From torchvision's Vgg-13. |
Vgg16 | TorchVision | torchvision's Vgg-16 | From torchvision's Vgg-16. |
Vgg19 | TorchVision | torchvision's Vgg-19 | From torchvision's Vgg-19. |