Skip to content

Latest commit

 

History

History
45 lines (37 loc) · 1.56 KB

README.md

File metadata and controls

45 lines (37 loc) · 1.56 KB
Raw

SimpleDG

This is the training and test code of our method for NICO challenge track1 and track2.

We won both the 2nd prize in track1 and track2, the paper has been accepted in ECCV2022 workshop. certificates

preperation

Download the NICO++ dataset HERE, unzip them to $data_dir and fill it to Makefile (line 1: data_dir = $data_dir). The data should have the following structure:

$data_dir
├── track_1  
│   ├── dg_label_id_mapping.json  
│   └── public_dg_0416  
│       ├── public_test_flat  
│       └── train  
├── track_2  
│   ├── ood_label_id_mapping.json  
│   └── public_ood_0412_nodomainlabel  
│       ├── public_test_flat  
│       └── train

train previous DG algorithms

  • train a single algorithm with specific parameters
make train algorithm=ERM
  • train multiple algorithms with hyperparameter searching, change the value of $(algorithms) in Makefile with what you want
make sweep

show results

show and compare the results of different algorithms

make collect_results

train with our method using ddp

enter the ddp_training directory and follow the steps in README.md

Acknowledgment

We conduct experiments on the base of DomainBed