gperdrizet/autoencoders

DF2K_OST 128×128

Dataset Description

Preprocessed version of the DF2K dataset (DIV2K + Flickr2K) for training image autoencoders.

• Total images: 900
• Image size: 128×128 pixels
• Format: PNG (RGB)
• Source: DIV2K high-quality image dataset

Dataset Structure

df2k_ost_128/
├── img_00000.png
├── img_00001.png
├── ...
└── img_00899.png

Usage

from huggingface_hub import hf_hub_download
import os

# Download all images
for i in range(900):
    filename = f"img_{i:05d}.png"
    hf_hub_download(
        repo_id="gperdrizet/autoencoders",
        filename=filename,
        repo_type="dataset",
        local_dir="./data"
    )

Or use with the custom data loader:

from src.data_utils import load_df2k_ost

images = load_df2k_ost(image_size=128)

Original Dataset

• DIV2K: https://data.vision.ee.ethz.ch/cvl/DIV2K/
• License: https://data.vision.ee.ethz.ch/cvl/DIV2K/

Preprocessing

1. Downloaded DIV2K train + validation sets
2. Resized to 128×128 using Lanczos resampling
3. Saved as optimized PNG files

Citation

If you use this dataset, please cite the original DIV2K paper:

@InProceedings{Agustsson_2017_CVPR_Workshops,
    author = {Agustsson, Eirikur and Timofte, Radu},
    title = {NTIRE 2017 Challenge on Single Image Super-Resolution: Dataset and Study},
    booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month = {July},
    year = {2017}
}