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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Unlabeled 3D objects present an opportunity to leverage pretrained visionlanguage models (VLMs) on a range of annotation tasks -- from describing objectsemantics to physical properties. An accurate response must take into accountthe full appearance of the object in 3D, various ways of phrasing thequestion/prompt, and changes in other factors that affect the response. Wepresent a method to marginalize over any factors varied across VLM queries,utilizing the VLM's scores for sampled responses. We first show that thisprobabilistic aggregation can outperform a language model (e.g., GPT4) forsummarization, for instance avoiding hallucinations when there are contrastingdetails between responses. Secondly, we show that aggregated annotations areuseful for prompt-chaining; they help improve downstream VLM predictions (e.g.,of object material when the object's type is specified as an auxiliary input inthe prompt). Such auxiliary inputs allow ablating and measuring thecontribution of visual reasoning over language-only reasoning. Using theseevaluations, we show how VLMs can approach, without additional training orin-context learning, the quality of human-verified type and materialannotations on the large-scale Objaverse dataset.

Evaluating VLMs for Score-Based, Multi-Probe Annotation of 3D Objects