glennwuwu/tiktok-techjam-2026-eval

TikTok TechJam 2026 Eval

Held-out demonstration pair used by Seer: COCO val2017 photographs plus the WildFake DALL·E Advanced (DALL·E 3) subset.

Do not train on this split.

Contents

label	meaning	count	origin
0 / real	photograph	5,000	COCO val2017
1 / fake	AI-generated	8,843	WildFake DALL·E Advanced

Columns: image, label, source, generator, id.

id is the COCO stem for reals, and {session}_{stem} for fakes so duplicate WildFake basenames stay unique.

Load

from datasets import load_dataset

ds = load_dataset("glennwuwu/tiktok-techjam-2026-eval", split="test")

Attribution / licence

This repo redistributes two existing collections for eval convenience.

• COCO val2017 — images keep their original Flickr licences; annotations are CC BY 4.0.
• WildFake DALL·E — from hy2628982280/WildFake on ModelScope (re-uploader tag apache-2.0). Only the Advanced / DALL·E 3 subset is included, not DALL·E 2 Typical.