Infoblock alpha
cs-cv/vision-language IB-arxiv-2607.28269-caption-recall measured

Automated image-blind LLM-as-a-Judge evaluation of the generated Incidents1M captions measured a recall of 46.0%.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

recall_pct 46.0

01

Primary source

  1. arxiv · 2607.28269 · v1 · abstract
Title
Theia: Large-Scale Multimodal Captioning and Automated Validation of the Incidents1M Dataset for Data-Free Distillation
Authors
Simone Giano, Lorenzo Severini, Alessandro Galdelli, Adriano Mancini
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
low Recall (46.0%)

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.