AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

VDC-Agent: When Video Detailed Captioners Evolve Themselves via Agentic Self-Reflection

Qiang Wang, XINYUAN GAO, SongLin Dong, Jizhou Han, Jiangyang Li, Yuhang He, Zhiheng Ma, Yihong Gong

ECCV 2026 · Multi-agent systems and games › LLM agents · approach: Self-evolving language agent learning · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add principle text to prompt (ablation)Qwen2.5-VL-7B · VDCaverage accuracy: 43.95 -> 46.47Improved
add test-time self-reflection on top of principles (ablation)Qwen2.5-VL-7B · VDCaverage accuracy: 46.47 -> 47.10Improved
add test-time self-reflection on top of principles (ablation)Qwen2.5-VL-7B · VDCinference time: 22.3s -> 164.9sHurt
internalize reflection by curriculum DPO training instead of test-time reflection (ablation)Qwen2.5-VL-7B · VDCaverage accuracy: 47.10 -> 49.08Improved
use curriculum (easy-to-hard) DPO instead of random order (ablation)VDC-Agent · VDCaverage accuracy: 48.03 -> 49.08Improved
use hard-to-easy order instead of random order (ablation)VDC-Agent · VDCaverage accuracy: 48.03 -> 47.65Hurt
use DPO curriculum instead of SFT (ablation)VDC-Agent · VDCaverage accuracy: 47.54 -> 49.08Improved
raise max iterations T from 4 to 6 (ablation)VDC-Agent · VDCaverage accuracy: 49.08 -> 49.55No clear effect
raise max iterations T from 2 to 4 (ablation)VDC-Agent · DREAM-1KF1: 34.0 -> 37.4Improved
swap principle set P1 for P2-P4 written by others (ablation)VDC-Agent · VDCaverage accuracy: 49.08 vs 48.84, 49.02, 48.95No clear effect
apply VDC-Agent to LLaVA-OneVision-7BLLaVA-OneVision-7B · VDCaverage accuracy: 37.45 -> 43.18Improved
apply VDC-Agent to InternVL2.5-8BInternVL2.5-8B · DREAM-1KF1: 27.6 -> 33.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy