Explainable Deepfake Detection Challenge
Paper • 2607.21007 • Published
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
ACM MM Explainability Challenge 2026 — MSU Team Checkpoints
This repository contains the detector and explanation-generation checkpoints for “MSU Team at the Explainable Deepfake Detection Challenge 2026: Grounded Artifact Evidence for Deepfake Detection”.
The pipeline combines image-level deepfake detection and Artifact Evidence Maps with class-conditional complex explanation generation and text-only explanation simplification.
Checkpoints
| Path | Description | Size |
|---|---|---|
detector.ckpt |
DINOv3 + Mesorch detector predicting the image-level fake probability and an Artifact Evidence Map. | 10.67 GB |
qwen/Qwen3-VL-8B_complex_fake/ |
Complex explanation generator for images classified as fake. | 17.55 GB |
qwen/Qwen3-VL-8B_complex_real/ |
Complex explanation generator for images classified as real. | 17.55 GB |
qwen/Qwen3-VL-8B_complex_to_simple/ |
Shared text-only simplifier converting complex explanations into simple explanations for both classes. | 17.55 GB |
The detector checkpoint contains complete model weights in the xplain-detector-v2 format. Each Qwen directory contains a complete model, configuration files, tokenizer resources, and a chat template.
Total download size: approximately 63.31 GB.
Links
- Source code
- Solution paper: Coming soon.
- Challenge paper
- XPlainVerse paper
- Challenge website
- Challenge leaderboard
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support