Spaces:
Running on Zero
Running on Zero
File size: 1,101 Bytes
130fc25 c7c1d7e 130fc25 c7c1d7e 130fc25 c7c1d7e 130fc25 c7c1d7e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | ---
title: AIGC Image Detector (TechJam 2026 Track 5)
emoji: 🔍
colorFrom: red
colorTo: blue
sdk: gradio
sdk_version: 6.26.0
python_version: "3.12"
app_file: app.py
pinned: false
license: mit
short_description: AI-generated image detector with evidence maps
---
# AIGC image detector — TechJam 2026 Track 5
Live demo of the detector from
[github.com/samt41/techjam_aigsee](https://github.com/samt41/techjam_aigsee):
a frozen DINOv2-B/14 backbone read at blocks 8 and 10 by two small transformer heads
whose logits are averaged. Upload, paste, or pick a sample; you get the probability the
image is AI-generated plus a signed evidence map (red = fake evidence, blue = real).
Headline numbers on the full track demonstration set: clean AUC 0.9978, robust mean
over six corruption families 0.9943, track score 0.9960. Full tables, error analysis
and training code are in the repo.
Sample images: AI-generated ones are from the Community Forensics dataset
(Park & Owens, CC-BY-4.0); real photographs are CC0 from Wikimedia Commons. Credits
in `samples/ATTRIBUTION.md`. This Space runs on ZeroGPU.
|