You can not select more than 25 topics
Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
lzqoscar
489287b7b7
|
3 months ago | |
|---|---|---|
| mindcv | 3 months ago | |
| README.md | 3 months ago | |
| amt_b_16_224_finetune_ascend.yaml | 3 months ago | |
| amt_b_16_224_pretrain_ascend.yaml | 3 months ago | |
| config.py | 3 months ago | |
| finetune.png | 3 months ago | |
| finetune.py | 3 months ago | |
| pretrain.png | 3 months ago | |
| pretrain.py | 3 months ago | |
README.md
Mindspore Implementation of the paper Good helper is around you: Attention-driven Masked Image Modeling
This is a Mindspore Implementation version for the work. Due to the features of mindspore, the training time for AMT is longer than pytorch version. For more details, please refer to the paper.
Our code is based on mindcv. Installation and preparation follow that repo.
Pretrain
python pretrain.py -c amt_b_16_224_pretrain_ascend.yaml
Finetune
python finetune.py -c amt_b_16_224_finetune_ascend.yaml
Loss log for pretrain
Accuracy for tiny-imagenet
