The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
RoSeDiT samples
The 50,000-sample archives behind the two headline numbers of Role-Separated Diffusion Transformers: Global Rule Isolation and Deterministic Sampling Manifold Compression (Zhiguang Liu, Yi Shang; University of Missouri), so these numbers can be recomputed without a GPU. Paper: https://arxiv.org/abs/2611.19009. Code: https://github.com/lz7fd/RoSeDiT. Checkpoints: https://huggingface.co/lz7fdmu/RoSeDiT
| File | Model | Steps | FID | sFID | IS | Prec. | Rec. |
|---|---|---|---|---|---|---|---|
RoSeDiT-XL-2_1400K_samples50k.npz |
RoSeDiT-XL/2 | 1.4M | 7.98 | 5.30 | 129.1 | 0.696 | 0.660 |
RoSeDiT-XL-2_DSMC_2400K_samples50k.npz |
RoSeDiT-XL/2 + DSMC | 2.4M | 5.84 | 4.67 | 150.4 | 0.708 | 0.666 |
RoSeDiT-XL/2 at 1.4M steps surpasses SiT-XL/2's 7M-step FID of 8.3 with 7.98. With DSMC, RoSeDiT-XL/2 at 2.4M steps surpasses SiT + REPA's 4M-step FID of 5.9 with 5.84.
Each file is a NumPy archive with one array arr_0 of shape (50000, 256, 256, 3), dtype uint8, the format of the
ADM evaluation suite. Each archive is about
9.2 GB; MD5SUMS lists the checksums.
The images are class-conditional ImageNet 256x256 samples from the EMA weights, generated by generate.py of the code
repository with a 250-step Euler-Maruyama SDE sampler, no classifier-free guidance, random class labels, seed 0, one
process and a batch size of 64.
Recompute a number
hf download lz7fdmu/RoSeDiT_samples RoSeDiT-XL-2_DSMC_2400K_samples50k.npz --repo-type dataset --local-dir .
python evaluator.py VIRTUAL_imagenet256_labeled.npz RoSeDiT-XL-2_DSMC_2400K_samples50k.npz
evaluator.py and the reference batch VIRTUAL_imagenet256_labeled.npz are those of the ADM evaluation suite.
Citation
@article{liu2026rosedit,
title = {Role-Separated Diffusion Transformers: Global Rule Isolation and Deterministic Sampling Manifold Compression},
author = {Liu, Zhiguang and Shang, Yi},
journal = {arXiv preprint arXiv:2611.19009},
year = {2026}
}
- Downloads last month
- 1