English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

When an object detector is deployed in a novel setting it often experiences adrop in performance. This paper studies how an embodied agent can automaticallyfine-tune a pre-existing object detector while exploring and acquiring imagesin a new environment without relying on human intervention, i.e., a fullyself-supervised approach. In our setting, an agent initially learns to explorethe environment using a pre-trained off-the-shelf detector to locate objectsand associate pseudo-labels. By assuming that pseudo-labels for the same objectmust be consistent across different views, we learn the exploration policy LookAround to mine hard samples, and we devise a novel mechanism calledDisagreement Reconciliation for producing refined pseudo-labels from theconsensus among observations. We implement a unified benchmark of the currentstate-of-the-art and compare our approach with pre-existing explorationpolicies and perception mechanisms. Our method is shown to outperform existingapproaches, improving the object detector by 6.2% in a simulated scenario, a3.59% advancement over other state-of-the-art methods, and by 9.97% in the realrobotic test without relying on ground-truth. Code for the proposed approachand baselines are available athttps://iit-pavis.github.io/Look_Around_And_Learn/.

Look Around and Learn: Self-Training Object Detection by Exploration