AI
EgoTools: Towards Tool-Centric Reasoning in Real-World Egocentric Videos
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 03/10/2026, 03:46:41 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 09:18:24 UTC.
Briefings
EgoTools introduces a comprehensive suite for egocentric tool-use understanding, comprising a 100-hour video corpus and a 1,000-question diagnostic benchmark. Experimental results show that while current models like Gemini-3.1-Pro achieve 66.9% overall accuracy, they struggle significantly with tool grounding, reaching only 51.7% on the Perception & Grounding track. Supervised fine-tuning on the benchmark improved Qwen3-VL-8B-Instruct from 50.0% to 60.9% accuracy under source-video separation.
Key points
- EgoTools consists of two components: EgoTools-Data, a 100-hour corpus of tool-centric egocentric recordings with synchronized audio and dense captions, and EgoTools-Bench, a 1,000 QA benchmark covering four tracks from perception to causal reasoning.
- Current models like Gemini-3.1-Pro achieve 66.9% overall accuracy on the benchmark but fail to ground tool use in visual evidence, with only 51.7% accuracy on the Perception & Grounding track.
Context and limitations
The study relies on a single preprint from September 2026, with no independent verification or long-term validation of the EgoTools-Data corpus or benchmark performance.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 03/10/2026, 03:46:41 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeibnvqeozehsc6b2tugnxtdrcbjwhh2iv2qfwnv2usmnswinkuirwa