Skip to content

Latest commit

 

History

History
84 lines (64 loc) · 1.52 KB

DATASET.md

File metadata and controls

84 lines (64 loc) · 1.52 KB
Raw

How to install datasets

We suggest putting all datasets under the same foulder of $DATASETS, such path is set in configs.default.py. The file structure looks like

$DATASETS/
|-- Office31/
|-- OfficeHome/
|-- VisDa2017/
|-- domain_net/

If you have some datasets already installed somewhere else, you can create symbolic links in $DATASETS/dataset_name that point to the original data to avoid duplicate download.

Office31

  • Create a folder named Office31 under $DATASETS.
  • Prepare domains following
Office31/
|-- amazon/
|-- dslr/
|-- webcam/
amazon.txt
dslr.txt
webcam.txt
  • Copy txts/Office31/*.txt to the current folder.

OfficeHome

  • Create a folder named OfficeHome under $DATASETS.
  • Prepare domains following
OfficeHome/
|-- Art/
|-- Clipart/
|-- Product/
|-- RealWorld/
Art.txt
Clipart.txt
Product.txt
RealWorld.txt
  • Copy txts/OfficeHome/*.txt to the current folder.

VisDa2017

  • Create a folder named VisDa2017 under $DATASETS.
  • Prepare domains following
VisDa2017/
|-- train/
|-- validation/
syn.txt
real.txt
  • Copy txts/VisDa2017/*.txt to the current folder.

domain_net

  • Create a folder named domain_net under $DATASETS.
  • Prepare domains following
domain_net/
|-- painting/
|-- real/
|-- sketch/
painting_train.txt
real_train.txt
sketch_train.txt
  • Copy txts/domain_net/*.txt to the current folder.

Custom dataset

Building your Custom dataset is straightforward by inheriting the class of datasets.benchmark.MultiDomainsBenchmark