Skip to content

YOLO-World

Real-time object detection with open vocabulary

Visit official site

What YOLO-World is

YOLO-World is described as a real-time open-vocabulary object detector built on the You Only Look Once detector series. It adds vision-language modeling and large-scale dataset pre-training to strengthen open-vocabulary detection, with RepVL-PAN and region-text contrastive loss connecting visual and language information.

What you can do with it

  • Run real-time open-vocabulary object detection
  • Detect varied objects in a zero-shot manner
  • Carry out zero-shot inference on the LVIS dataset

Model notes

  • Build on the YOLO series with vision-language modeling and large-scale pre-training
  • Use RepVL-PAN and region-text contrastive loss for visual-language interaction
  • State 35.4 AP and 52.0 FPS on LVIS with a V100
  • Fine-tuned versions cover object detection and open-vocabulary instance segmentation

Pros & cons

✓ What we like

  • High efficiency and high speed are described
  • Fine-tuned versions serve several downstream tasks

! What to watch out for

  • Open-vocabulary detection relies on vision-language pre-training
  • The stated LVIS figures are measured on a V100

FAQ

What does it detect?

The material describes real-time open-vocabulary detection of varied objects in a zero-shot manner.

Which techniques are named?

Vision-language modeling, large-scale pre-training, RepVL-PAN and region-text contrastive loss.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review