FindIt: A Format-Informed Visual Detection Benchmark for Generalist Multimodal LLMs

Name: FindIt: A Format-Informed Visual Detection Benchmark for Generalist Multimodal LLMs
Uploaded: 2026-10-07T22:53:43.806257+00:00
Duration: 333 s
Description: This work created a comprehensive benchmark to evaluate how well generalist multimodal models localize objects using text prompts across four tasks: object detection, referring expression detection, instance localization, and video object detection. It found that model performance is highly dependent on the output format requested in the prompt.

This work created a comprehensive benchmark to evaluate how well generalist multimodal models localize objects using text prompts across four tasks: object detection, referring expression detection, instance localization, and video object detection. It found that model performance is highly dependent on the output format requested in the prompt.

Open the full episode