Classification Experiments
September 9, 2021 · View on GitHub
Dataset Preparation
ImageNet-LT
- Download ImageNet Full dataset
- ImageNet-LT list and frequency file:
# train/val/test split
https://drive.google.com/drive/folders/1j7Nkfe6ZhzKFXePHdsseeeGI877Xu1yf
# Frequency File
wget https://github.com/Megvii-BaseDetection/DisAlign/releases/download/LVIS/imagenet_lt_category_frequency.json
- Put the data as following:
cvpods/datasets/imagenetlt
├── imagenet_lt_category_frequency.json
├── ImageNet_LT_test.txt
├── ImageNet_LT_train.txt
├── ImageNet_LT_val.txt
├── train/
└── val/
iNaturalist2018
Places-LT
Model Zoo
ImageNet-LT
| Name | Cls Norm | Input Size | Epoch | Top-1 Acc(Test) | Top-5 Acc(Test) | Trained Model |
|---|---|---|---|---|---|---|
| ResNet-50 | 224 | 90 | LINK(TODO) | |||
| ResNet-50 | Cosine | 224 | 90 | LINK(TODO) | ||
| ResNeXt-50 | 224 | 90 | 46.918 | 73.088 | LINK | |
| ResNeXt-50 | Cosine | 224 | 90 | LINK(TODO) | ||
| ResNet-50+DisAlign | 224 | 90 | LINK(TODO) | |||
| ResNet-50+DisAlign | Cosine | 224 | 90 | LINK(TODO) | ||
| ResNeXt-50+DisAlign | 224 | 90 | 52.438 | 77.068 | LINK | |
| ResNeXt-50+DisAlign | Cosine | 224 | 90 | LINK(TODO) |