search

yorkeyao / VehicleX

VehicleX: Simulating Content Consistent Vehicle Datasets with Attribute Descent (ECCV 2020, TPAMI 2023)
156 stars 19 forks source link

about style domain adaptation #11

Closed sezarxray closed 3 years ago

sezarxray commented 3 years ago

Hi, what a nice project! Could you plz tell me more about training data for style domain adaptation? what is trainA/B? what is testA/B? How many data we need?

│   └───sys2real
|   │   │ trainA
|   │   │ trainB
|   │   │ testA
|   │   │ testA

Thanks in advance

yorkeyao commented 3 years ago

Thank you for your question.

For training, if you want to perform a style transfer from VehicleX to VeRi. VehicleX is trainA and the training set of VeRi is trainB. Then you will get a model Gb to generate VeRi style VehicleX data.

In my setting, I may not really need a test set. Thank you for pointing this out and I have modified the README. You may need to modify the args source_path and save_path in test_spgan.py to inference. In this case, the source_path is the path to VehicleX data.

sezarxray commented 3 years ago

How many data we need?

yorkeyao commented 3 years ago

I use the entire VehicleX training set and entire VeRi training set.

sezarxray commented 3 years ago

thx