NICO

Introduction

🤗 We achieved third place 🏆 in the 2022 NICO Common Context Generalization Challenge (ECCV 2022 Workshop), and the related code will be released here.

Datasets:

(NICO++)[https://arxiv.org/abs/2204.08040]

downloads:

The released data (for NICO challenge) is available:

dropbox

Tsinghua Cloud

make ensure that the files are placed as follows:

+-NICO/
  |
  *-dg_label_id_mapping.json
  |
  +-nico/
    |
    +-test/
    |
    +-train/
      |
      +-rock/
      | |
      | +-hot air balloon/
      | |
      | +-goose/
      | |
      | +-frog/
      | |
      | +-mailbox/
      | |
      | +-bus/
      | |
      | +-shrimp/
      | |
      | +-airplane/
      | |
      | +-crocodile/
      | |
      | +-pineapple/
      | |
      | +-cow/
      | |
      | +-tortoise/
      | |
      | +-sheep/
      | |
      | +-scooter/
      | |
      | +-lion/
      | |
      | +-seal/
      | |
      | +-dolphin/
      | |
      | +-pumpkin/
      | |
      | +-racket/
      | |
      | +-fox/
      | |
      | +-sunflower/
      | |
      | +-car/
      | |
      | +-corn/
      | |
      | +-elephant/
      | |
      | +-sailboat/
      | |
      | +-dog/
      | |
      | +-tent/
      | |
      | +-flower/
      | |
      | +-football/
      | |
      | +-hat/
      | |
      | +-chair/
      | |
      | +-cat/
      | |
      | +-owl/
      | |
      | +-cactus/
      | |
      | +-fishing rod/
      | |
      | +-ship/
      | |
      | +-clock/
      | |
      | +-wheat/
      | |
      | +-spider/
      | |
      | +-umbrella/
      | |
      | +-horse/
      | |
      | +-ostrich/
      | |
      | +-giraffe/
      | |
      | +-wolf/
      | |
      | +-helicopter/
      | |
      | +-kangaroo/
      | |
      | +-bicycle/
      | |
      | +-bird/
      | |
      | +-butterfly/
      | |
      | +-motorcycle/
      | |
      | +-monkey/
      | |
      | +-rabbit/
      | |
      | +-crab/
      | |
      | +-squirrel/
      | |
      | +-bear/
      | |
      | +-train/
      | |
      | +-tiger/
      | |
      | +-lifeboat/
      | |
      | +-lizard/
      | |
      | +-truck/
      | |
      | +-gun/
      |
      +-outdoor/
      | |
      | +-hot air balloon/
      | |
      | +
      | .
      | .
      +-autumn/
      | |
      | +-hot air balloon/
      | |
      | +
      | .
      | .
      +-dim/
      | |
      | +-hot air balloon/
      | |
      | +
      | .
      | .
      +-water/
        |
        +-hot air balloon/
        |
        +
        .
        .

You can also free to use NICO++ data for your research for non-economic purpose.

The most important thing you should pay attention to is:

Make sure the last character of the root directory of all image folder's path is ''

How to train track1 dataset immediately?

1. Create your environment for training

conda env create -f environment.yaml
source activate nico-mcislab840 # in [Linux] , activate nico-mcislab840 # in [Window]

2. Modify train_image_path, label2id_path, and test_image_path in track1_run.sh

3. Run ensemble_track1_run.sh for ensemble, and then get three final checkpoints: `track_1_pth_1.pth`, `track_1_pth_2.pth`, `track_1_pth_3.pth`

chmod 777 ensemble_track1_run.sh
bash ensemble_track1_run.sh

How to test track1 dataset immediately?

1. Modify test_image_path, label2id_path, and test_pth_path in test.sh

2. Run test.sh for Test Time Augmentation (a longer period of time is required).

chmod 777 test.sh
bash test.sh

3. The voting method is applied for the final prediction (final_prediction.json)

python ensemble.py --ensemble_path predictionnico1 --save_path final_prediction.json

shaoshitong/NICO