Research map / Video / Video-language understanding
LongEgoRefer: A Benchmark for Long-Form Egocentric Video Referring Expression Comprehension
Shunya Kato, Taiki Miyanishi, Shuhei Kurita, Mahiro Ukai, Nakamasa Inoue, Chenhui Chu
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 |
|---|---|---|---|
| increase max input frames from 64 to 500 (ablation) | GPT-5 · LongEgoRefer (Ego4D, 1,498 queries) | tIoU@0.5: 3.60 -> 12.55 | Improved |
| increase reasoning effort from minimal to high (ablation) | GPT-5 · LongEgoRefer (Ego4D, 1,498 queries) | tIoU@0.5: 8.7 -> 12.61 | Improved |
| increase reasoning effort from medium to high (ablation) | GPT-5 · LongEgoRefer (Ego4D, 1,498 queries) | tIoU@0.5: 12.55 -> 12.61 | No clear effect |
| use oracle temporal segmentation instead of GPT-5 temporal grounding (ablation) | Grounded SAM2 · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 7.89 -> 43.54 | Improved |
| use oracle first-frame box prompt with SAM2.1 Base instead of Grounding DINO box (ablation) | SAM2.1 · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 43.54 -> 67.21 | Improved |
| use SAM2.1 Large instead of Base with oracle prompt (ablation) | SAM2.1 · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 67.21 -> 67.82 | No clear effect |
| replace Grounded-SAM2 with SAM3 as spatial stage (ablation) | GPT-5 pipeline · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 7.89 -> 3.86 | Hurt |
| replace Grounded-SAM2 with SAM3 as spatial stage (ablation) | Gemini 2.5 Pro pipeline · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 5.49 -> 2.23 | Hurt |
| evaluate on 6-minute EgoTracks clips instead of full 45-minute videos | Gemini 2.5 Pro · EgoTracks vs Ego4D | mtIoU: +12.24 mtIoU | Improved |
| scale from Gemini 2.5 Flash to Gemini 2.5 Pro | Gemini · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 2.21 -> 5.49 | Improved |
| increase target appearance rate from (0,1] to (2,25] (ablation) | GPT-5 · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 5.16 -> 13.54 | Improved |
| lengthen referring expression from (0,50] to (70,110] words (ablation) | GPT-5 · LongEgoRefer (Ego4D, 1,498 queries) | mvIoU: 10.29 -> 6.05 | Hurt |
Closest papers by meaning
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.88)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.88)
- A Large-Scale Human-Centric Benchmark for Referring Expression Comprehension in the LMM Era (NeurIPS 2024 · similarity 0.88)
- MMEgo: Towards Building Egocentric Multimodal LLMs for Video QA (ICLR 2025 · similarity 0.87)
- VideoRefer Suite: Advancing Spatial-Temporal Object Understanding with Video LLM (CVPR 2025 · similarity 0.87)
- Omnia de EgoTempo: Benchmarking Temporal Understanding of Multi-Modal LLMs in Egocentric Videos (CVPR 2025 · similarity 0.87)
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.
