/TECH

Perceptron Egocentric Beats Gemini Pipeline by 77% F1 on WGO-Bench

AA#600

Combined views

356

2 posts, first seen 4h ago

4 likes2 comments

Source Posts on X

Cluster sources

Armen Aghajanyan@ArmenAghaTECH

Combined views

356

2 posts, first seen 4h ago

4 likes

1 reposts

Sentiment

Positive0%0%Negative

Guardrails removed spam, off-topic, unclear, or duplicate replies.

Digg Deeper

Ask a question below.

Published answers will appear here.

Source Posts on X

Cluster sources

Armen Aghajanyan@ArmenAghaTECH

4/ Built on Mk1, our frontier perceptive model, currently SOTA on a bulk of embodied reasoning benchmarks. Nothing here was fine-tuned for WGO-Bench. The capability is the model: parse physical activity, segment it into actions, and return supervision we can train on. https://x.com/ArmenAgha/status/2075262932695609410/photo/1

Armen Aghajanyan@ArmenAghaTECH

3/ On WGO-Bench, Perceptron Egocentric is +77% relative semantic end-to-end F1 over the prior best pipeline (built on Gemini series of models). It is also 10-15x cheaper than human annotation. https://x.com/ArmenAgha/status/2075262930745213286/photo/1

4/ Built on Mk1, our frontier perceptive model, currently SOTA on a bulk of embodied reasoning benchmarks. Nothing here was fine-tuned for WGO-Bench. The capability is the model: parse physical activity, segment it into actions, and return supervision we can train on. https://x.com/ArmenAgha/status/2075262932695609410/photo/1

Armen Aghajanyan@ArmenAghaTECH

3/ On WGO-Bench, Perceptron Egocentric is +77% relative semantic end-to-end F1 over the prior best pipeline (built on Gemini series of models). It is also 10-15x cheaper than human annotation. https://x.com/ArmenAgha/status/2075262930745213286/photo/1

2 comments

1 reposts

Sentiment

Positive0%0%Negative

Guardrails removed spam, off-topic, unclear, or duplicate replies.

Digg Deeper

Ask a question below.

Published answers will appear here.