Research map / Vision-language and multimodal / Multimodal LLMs
EgoPlan-Bench2: A Benchmark for Multimodal Large Language Model Planning in Real-World Scenarios
Lu Qiu, Yi Chen, Yuying Ge, Yixiao Ge, Ying Shan, Xihui Liu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add ground-truth action sequence prompt (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 51.67(+18.87) | Improved |
| add GPT-generated action sequence prompt (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 36.71(+3.91) | Improved |
| add video-level description prompt (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 32.45(-0.35) | No clear effect |
| add frame-level descriptions (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 32.91(+0.11) | No clear effect |
| add key object tracking (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 31.88(-0.92) | Hurt |
| add object bounding boxes (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 37.63(+4.83) | Improved |
| add hand bounding boxes (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 36.82(+4.02) | Improved |
| add object status prompt (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 29.46(-3.34) | Hurt |
| add scene graph with object, attribute and relation (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 32.80 -> 31.88(-0.92) | Hurt |
| add CoT reasoning to object bounding boxes (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 37.63 -> 39.82(+7.02) | Improved |
| add action sequence to bounding box with CoT (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 39.82 -> 42.81(+10.01) | Improved |
| add self-consistency to CoT pipeline (ablation) | GPT-4V · EgoPlan-Bench2 (869 QA subset), GPT-4V prompting | accuracy: 42.81 -> 43.04(+10.24) | Improved |
Closest papers by meaning
- On Path to Multimodal Generalist: General-Level and General-Bench (ICML 2025 · similarity 0.91)
- MDK12-Bench: A Multi-Discipline Benchmark for Evaluating Reasoning in Multimodal Large Language Models (AAAI 2026 · similarity 0.89)
- LLMs Can Plan Only If We Tell Them (ICLR 2025 · similarity 0.89)
- SEED-Bench: Benchmarking Multimodal Large Language Models (CVPR 2024 · similarity 0.88)
- MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark (ICML 2024 · similarity 0.88)
- TravelPlanner: A Benchmark for Real-World Planning with Language Agents (ICML 2024 · similarity 0.88)
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.
