Official implementation for TransDA

Official pytorch implement for “Transformer-Based Source-Free Domain Adaptation”.

Overview:

Official-implementation-for-TransDA

Result:

result_office31

result_officehome

Prerequisites:

  • python == 3.6.8

  • pytorch ==1.1.0

  • torchvision == 0.3.0

  • numpy, scipy, sklearn, PIL, argparse, tqdm

Prepare pretrain model

We choose R50-ViT-B_16 as our encoder.

wget https://storage.googleapis.com/vit_models/imagenet21k/R50+ViT-B_16.npz 
mkdir ./model/vit_checkpoint/imagenet21k 
mv R50+ViT-B_16.npz ./model/vit_checkpoint/imagenet21k/R50+ViT-B_16.npz

Our checkpoints could be find in Dropbox

Dataset:

  • Please manually download the datasets Office, Office-Home, VisDA, Office-Caltech from the official websites, and modify the path of images in each ‘.txt’ under the folder ‘./data/’.

  • The script "download_visda2017.sh" in data fold also can use to download visda

Training

Office-31

```python
sh run_office_uda.sh


### Office-Home


1
sh run_office_home_uda.sh


### Office-VisDA


1
sh run_visda.sh


# Reference



ViT




TransUNet




SHOT




## GitHub



https://github.com/ygjwd12345/TransDA



                                
                                    
Source: https://pythonawesome.com/official-implementation-for-transda/