Home

PP-MisDet Workshop @ CVPR 2026

DGM⁴+ is a benchmark and challenge for detecting semantic and narrative inconsistencies in image–caption pairs, with a focus on global scene plausibility and multimodal misinformation.

---

Motivation

Modern misinformation increasingly relies on contextual manipulation, where realistic foregrounds are combined with misleading backgrounds and deceptive captions.

DGM⁴+ moves beyond low-level artifact detection and emphasizes scene-level reasoning and narrative coherence.

---

Main Sections

• 📊 Dataset — Dataset design, generation, and statistics
• 🏁 Challenge — Tasks, submission, evaluation, timeline
• 📥 Download — Data access
• 📈 Baselines — Reference models
• ❓ FAQ — Common questions

---

Reference

Paper:
DGM⁴+: Dataset Extension for Global Scene Inconsistency Detection
https://arxiv.org/abs/2509.26047