---
dataset_info:
  features:
  - name: source_image
    dtype: image
  - name: target_image
    dtype: image
  - name: q
    dtype: int64
  - name: resample
    dtype: int64
  splits:
  - name: train
    num_bytes: 827602804.136
    num_examples: 1024
  download_size: 827037492
  dataset_size: 827602804.136
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: unknown
size_categories:
- 10K<n<100K
---

# Manga panels

for upscaling / denoising tasks

## columns
- `target_image`: contains the original-quality image
- `source_image`: downscaled, lower-quality image
- `q`: jpeg quality used in compressing the `source_image`