AI Paper Atlas

Research map / Large language models / Agents, code and math

E-comIQ-ZH: A Human-Aligned Dataset and Benchmark for Fine-Grained Evaluation of E-commerce Posters with Chain-of-Thought

Meiqi Sun, Mingyu Li, Junxiong Zhu

CVPR 2026 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation · 1 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 domain-specific SFT on E-comIQ-18k training set (ablation)Qwen2.5-VL-7B · E-comIQ-18k test (overall)SRCC: 0.119 -> 0.346Improved
add GRPO stage with accuracy and distribution rewards after SFT (ablation)Qwen2.5-VL-7B+SFT · E-comIQ-18k test (overall)SRCC: 0.346 -> 0.433Improved
add GRPO stage with accuracy and distribution rewards after SFT (ablation)Qwen2.5-VL-7B+SFT · E-comIQ-18k test (background)SRCC: 0.530 -> 0.520No clear effect
add SFT before GRPO instead of GRPO only (ablation)Qwen2.5-VL-7B · E-comIQ-18k test (overall)SRCC: 0.154 -> 0.346Improved
add accuracy-only GRPO after SFT (ablation)Qwen2.5-VL-7B+SFT · E-comIQ-18k test (overall)SRCC: 0.346 -> 0.360Improved
add distribution reward term to the accuracy reward (ablation)SFT+GRPO · E-comIQ-18k test (overall)SRCC: 0.360 -> 0.410Improved
apply original Q-Insight GRPO reward (no CoT) on the datasetQ-Insight · E-comIQ-18k test (overall)SRCC: 0.152 -> 0.235Improved
use E-comIQ-M as automatic evaluator on out-of-domain benchmark vs human scoresE-comIQ-M · E-comIQ-BenchPLCC/SRCC with human: around 0.34Mixed
use OCR-based text metrics to evaluate Chinese text renderingE-comIQ-Bench · E-comIQ-BenchPhrase F1 vs human text score: GPT-4o phrase F1 0.86, human text 2.83Hurt

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