---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: mask
    dtype: image
  splits:
  - name: train
    num_bytes: 168179584.5
    num_examples: 4076
  - name: validation
    num_bytes: 239947212.14
    num_examples: 1020
  - name: test
    num_bytes: 171513671
    num_examples: 572
  download_size: 1618850322
  dataset_size: 579640467.64
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
task_categories:
- image-segmentation
size_categories:
- 1K<n<10K
---

# Link
* Paper: https://www.cs.cityu.edu.hk/~rynson/papers/cvpr20c.pdf
* Project page: https://jiaying.link/cvpr2020-pgd/

# Split info
We use the splitted version of the dataset. We split the train to train and validation with the ratio 80% and 20% respectively. If you want to use the original split composition, just combine train and validation.

# Dataset info
We found that there is a missing file in ADE20K test set. We found the missing file in the combined zip file in the project page.

# License info
Please refer to the paper and project page for the details.