Dataset / SID /Fake /README.md
Edward0707's picture
Initial commit
1d2e6f2
|
Raw History Blame Contribute Delete
2.05 kB
# SID/Fake
This prefix contains 27,380 images from SID_Set with
`label=1` (`fake`).
**These JPEG bytes are not the source bytes.** The source images are lossless
PNG, and shipping them as PNG next to JPEG real images would hand any classifier
the file format as a perfect shortcut. Each image was therefore encoded exactly
once from its PNG master, with the encoder settings drawn from the measured
distribution of `SID/real`: 2,431 distinct profiles over 26,131 real images,
allocated by exact largest-remainder proportion rather than by i.i.d. draws.
The resulting profile distribution sits at total variation 0.0061 from the real
side. Matching only the format would have moved the shortcut from "which
format" to "which quantization table", and that one is baked into the pixels.
## Selection
label=1; every SID_Set FLUX row; PNG master re-encoded to a profile drawn from the real side
- `flux`: 27,380
Total: 27,380 images.
## Dimensions
Every image is exactly 1024x1024, so the `width > 512 AND height > 512` filter
the `NTIRE/*` and `Complementary` prefixes apply would have kept all of them
unchanged. Train with native-scale random cropping rather than resizing --
resampling is itself a signal, and one this set would carry and the real side
would not.
## Load with Datasets
```python
from datasets import load_dataset
dataset = load_dataset(
"TechJam2026-Jamlai-Bench/TechJam2026_Jamlai_Bench",
data_files={"train": "SID/Fake/train-*.parquet"},
)
```
The Parquet schema is identical to the `NTIRE/*` prefixes: the same sixteen columns in
the same order, so the prefixes concatenate. Everything specific to this set --
generator, source archive, the JPEG profile actually applied -- lives in
`selection_manifest.csv` instead, which also records every source row, output Parquet
row, image dimensions, and SHA-256 digest. `summary.json` records the selection rule,
per-file checksums, and verification results.
Before making a public redistribution, verify that the source dataset license and
attribution terms permit it.