Datasetpublic

KonstantinosKK/reflect-dit-train-images

Reflect-DiT Training Images This is the official dataset repository for the training images used in Reflect-DiT, a Reflective Diffusion Transformer for image generation. 🔗 Paper: Reflect-DiT: Inference-Time Scaling for Text-to-Image Diffusion Transformers via In-Context Reflection Contents The dataset is stored in multiple .tar archives located in the data/ directory: data/ ├── gen_eval_sana_part_0.tar ├── gen_eval_sana_part_1.tar ├── ... └──… See the full description on the dataset page: https://huggingface.co/datasets/KonstantinosKK/reflect-dit-train-images.

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes92downloads

Dataset Card

Reflect-DiT Training Images

This is the official dataset repository for the training images used in Reflect-DiT, a Reflective Diffusion Transformer for image generation.

🔗 Paper: Reflect-DiT: Inference-Time Scaling for Text-to-Image Diffusion Transformers via In-Context Reflection

The dataset is stored in multiple .tar archives located in the data/ directory:

data/
├── gen_eval_sana_part_0.tar
├── gen_eval_sana_part_1.tar
├── ...
└── gen_eval_sana_part_9.tar

Usage

Due to the dataset's size, it is provided as .tar files. You can manually download and extract them:

bash

mkdir extracted_data
tar -xvf data/gen_eval_sana_part_0.tar -C extracted_data/

You can also extract the files programmatically using Python:

python

import tarfile
from pathlib import Path

tar_dir = Path("data")
for tar_file in sorted(tar_dir.glob("*.tar")):
    with tarfile.open(tar_file) as archive:
        archive.extractall(path="extracted_data/")

⚠️ The dataset is not directly loadable via the datasets library due to its format.

License

This dataset is released under the Apache 2.0 license.