SeFi-Image/SeFi-Image-SemVAE-Demo

SeFi-Image SemVAE Demo

Eight small, standalone synthetic images for testing the public SeFi-Image SemVAE loader and feature-compression path. This is a smoke-test fixture, not a training set or an evaluation benchmark.

Each sample is available twice:

• as an ordinary JPEG under images/, for path-based command-line examples;
• as embedded image bytes in the test Parquet split, for dataset-loader examples.

No contact sheets, plots, or images from the SeFi code repository's assets/ directory are included.

Quick use

Download one image for the SeFi demo CLI:

hf download SeFi-Image/SeFi-Image-SemVAE-Demo \
  images/city_taxi.jpg \
  --repo-type dataset \
  --local-dir outputs/demo_data/semvae

Or load all eight rows from Parquet:

from datasets import load_dataset

dataset = load_dataset(
    "SeFi-Image/SeFi-Image-SemVAE-Demo",
    split="test",
)
image = dataset[0]["image"]
prompt = dataset[0]["prompt"]

Data fields

The Parquet split contains:

• id, filename, image, width, height, and sha256;
• prompt, origin_caption, and enhanced_prompt from the upstream sample;
• pinned upstream repository, revision, subset, shard, and sample id;
• upstream style/category, prompt-generator, image-generator metadata, and the per-row license label;
• semvae_token_cosine, a checked deterministic reconstruction smoke metric.

The raw files are:

images/
├── city_taxi.jpg
├── future_plaza.jpg
├── garden_fountain.jpg
├── labrador_beach.jpg
├── mountain_meadow.jpg
├── overgrown_city.jpg
├── space_frigate.jpg
└── storm_sailboat.jpg

Provenance and license

The samples are repackaged without pixel modification from the synthetic_enhanced_prompt_random_resolution subset of ma-xu/fine-t2i at the pinned revision, whose dataset card declares Apache-2.0 and describes its synthetic images as generated by Z-Image and FLUX2. The real-image curated subset was deliberately excluded. source_image_generator preserves the per-row value (Z-Image-Turbo or FLUX.2-dev).

The exact upstream revision is 28fdd5663ee202b5cafc01d6ed08a03f14957854; all eight rows come from synthetic_enhanced_prompt_random_resolution/train-000000.tar. Image hashes were checked byte-for-byte against that pinned shard. Per-sample identifiers and hashes are recorded in manifest.json and in the Parquet rows.

origin_caption is the upstream .txt content. For these eight enhanced-prompt samples it is identical to enhanced_prompt; prompt retains the shorter original request from the upstream JSON.

The repackaged dataset is provided under Apache-2.0; see LICENSE. Please also cite the upstream Fine-T2I work when using these samples beyond a local smoke test.

@misc{ma2026finet2i,
  title={Fine-T2I: An Open, Large-Scale, and Diverse Dataset for High-Quality T2I Fine-Tuning},
  author={Xu Ma and Yitian Zhang and Qihua Dong and Yun Fu},
  year={2026},
  eprint={2602.09439},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}

Checked SemVAE result

Using the public dinov2_vitl14_reg/transformer_ch16 checkpoint, DINOv2-L/14 with registers, the documented preprocessing, and posterior mode, all eight images passed the 0.80 mean token-cosine threshold. Checked means range from 0.886340 to 0.922122. These numbers only detect obvious loading, preprocessing, token-slicing, and compression failures; they do not measure general model quality, and small hardware/version differences are expected.