Hugging Face has published a paper describing Ambient's submission to the EgoProactive track of the ECCV 2026 Wearable AI Challenge. According to the paper, the entry ranked first in the large-model division and second in the ≤2B division.
The task asks a wearable assistant to decide, after each eight-second segment of egocentric video, whether to intervene or stay silent. Ambient's approach has two main components.
First, the team reformulated intervention timing as single-token classification. Instead of generating either interrupt<utterance> or silent, the model predicts yes or no, and the decision is derived from the renormalised probabilities of those two tokens. The paper reports that this formulation improved macro-F1 by 0.249 and G-mean by 0.30 over free-form generation.
Second, because labelled data were limited to the released validation set, the team generated additional supervision using a tool-calling video agent that inspects each clip and assigns intervention timestamps. A narration-only alternative was four times larger and ten times cheaper, but transferred worse than supervision from an unrelated real corpus — a result the authors read as evidence that visual grounding matters more than annotation volume for this task.