The pre-trained diffusion model for Imagenet dataset (256x256_diffusion_uncond.pt) from guided-diffusion and CIFAR-10 dataset (unconditional CIFAR-10) from improved-diffusion.
You can find configs and checkpoints of recognition models in mmclassification. Specifically, we utilize the below three models for Imagenet dataset in our paper as follows:
| Model | Pretrain | Params(M) | Flops(G) | Top-1 (%) | Top-5 (%) | Config | Download |
|---|---|---|---|---|---|---|---|
| ResNet-50 | From scratch | 25.56 | 4.12 | 76.55 | 93.06 | config | model | log |
| Swin-T | From scratch | 28.29 | 4.36 | 81.18 | 95.61 | config | model | log |
| ConvNeXt-T | From scratch | 28.59 | 4.46 | 82.05 | 95.86 | config | model |
Moreover, we utilize the below two models for CIFAR-10 dataset in our paper as follows:
| Model | Pretrain | Params(M) | Flops(G) | Top-1 (%) | Config | Download |
|---|---|---|---|---|---|---|
| ResNet-18 | From scratch | 11.17 | 0.56 | 94.82 | config | model | log |
| ResNet-50 | From scratch | 23.52 | 1.31 | 95.55 | config | model | log |
Note: Please note that you need to copy recognition models to the respective folders in 'DistilledClassifier/saved/clean/IndTrainer//train/pretrained_model/'. For instance, resnet50 model with imagenet dataset will have <model-string> as convnextT_MMCV_Imagenet