Perceptron launches self-service API for egocentric video annotation at one-eighth the cost of GPT-6 Astra

Perceptron has released Egocentric 1.0, a self-service API for egocentric video annotation. On its upcoming EgoHome benchmark, the default 4 fps mode scores 0.787, within 0.010 of GPT-6 Astra at roughly one-eighth the cost; at 8 fps it remains about one-fifth the cost. The company says its 3D hand model reconstructs hand shape 12–25% more accurately than ACE-Ego-Hand on HOT3D and ARCTIC, and it reports the highest segment F1 (0.583) on WGO-Bench. It outperforms GPT-6 Astra on 0.5–1s subtasks (0.672 vs 0.649). EgoHome will be open-sourced; batches support up to 1,000 videos.

✨ AI summary of the linked source — not editorial content.

Explore the topic

• Launch
• API
• Video