54 upvotes
WeVisOne-VL merges object detection and segmentation in one model
What is WeVisOne-VL: Unified Visual Perception from Objects to Pixels?
WeVisOne-VL is a vision-language model that predicts the next token to describe both whole objects and individual pixels, letting it handle tasks from classification to segmentation with a single architecture. It uses a structured next-token prediction scheme to keep fine-grained visual details reliable.
Who it is for
Researchers and developers building multimodal apps can replace separate object-detectors and segmenters with a single model, simplifying pipelines and improving consistency.