Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
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.
AllenWrong
8f78085c13
|
1 year ago | |
|---|---|---|
| .gitignore | 1 year ago | |
| LICENSE | 1 year ago | |
| README.md | 1 year ago | |
| args.json | 1 year ago | |
| dataset.py | 1 year ago | |
| eval.sh | 1 year ago | |
| eval_linear.py | 1 year ago | |
| foundation_vit.py | 1 year ago | |
| lib.py | 1 year ago | |
| loss.py | 1 year ago | |
| models.py | 1 year ago | |
| train.py | 1 year ago | |
| train_linear.sh | 1 year ago | |
| train_vit.sh | 1 year ago | |
| transforms.py | 1 year ago | |
| utils.py | 1 year ago | |
| vision_transformer.py | 1 year ago | |
README.md
dino-paddle
Dino paddle implemetion
full_checkpoint 提取码：uyt7
linear_checkpoint 提取码：3j5t
使用
训练backbone
在train_vit.sh设置如下参数：
- --gpus：gpu ids
- --data_path：训练集路径
- --output_dir：日志、模型保存的目录
训练分类层
在train_linear.sh设置如下参数：
-
--gpus：gpu ids
-
--data_path：训练集路径
-
--pretrained_weights：backbone的预训练模型。默认即可
-
--output_dir：日志、模型的保存目录
-
--batch_size：32，训练分类层、评估模型时都是32。四卡的话就是4*32=128
评估模型
在eval.sh中设置如下参数：
- --data_path：验证集路径
- --pretrained_weights：backbone的预训练模型
- --pretrained_linear：分类层的预训练模型
No Description
Python Shell