LocateAnything: Fast and High-Quality Vision-Language Grounding with Parallel Box Decoding
Shihao Wang, Shilong Liu, Yuanguo Kuang, Xinyu Wei, Yangzhou Liu, Zhiqi Li, Yunze Man, Guo Chen, Andrew Tao, Guilin Liu, Jan Kautz, Lei Zhang, Zhiding Yu
arXiv:2605.27365 · 145▲ · cs.CV, cs.AI, cs.LG, cs.RO
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
LocateAnything is a vision-language model for object detection and grounding that uses Parallel Box Decoding (PBD), which predicts all four coordinates of a bounding box in a single forward step instead of generating them token by token. It treats each box (or point) as an atomic block with bidirectional intra-block attention, trained jointly with a standard next-token prediction objective and offering three inference modes (Fast, Slow, Hybrid).
Why it matters
Standard VLM grounding serializes each 2D box into a 1D token stream, so latency scales with the number of boxes and gets painful in dense scenes. PBD breaks that bottleneck, holding generation time roughly flat as box count grows, which makes VLM-based detection more viable for latency-sensitive uses like robotics and GUI agents. The Hybrid mode falls back to sequential decoding only when a parallel block looks malformed or spatially ambiguous, so you keep most of the speed without the accuracy cliff.
Practical takeaway
If you use a VLM to emit bounding boxes, watch for box-aligned parallel decoding showing up in grounding models as a way to cut inference cost in multi-object scenes. The authors report the same PBD design transfers to a Qwen3-VL-4B backbone, suggesting it is not tied to one architecture.
Key result
In the default Hybrid mode, LocateAnything-3B hits 12.7 boxes per second versus 5.0 for the comparable Rex-Omni-3B (2.5x) and about 1.1 for textual-coordinate Qwen3-VL (over 10x), while also improving mean F1 by +3.8 on LVIS and +1.8 on COCO. Caveat: the headline throughput numbers were all measured on COCO with batch size 1 on a single H100, and Fast mode alone drops accuracy in dense scenes (for example Dense200 F1 falls from 61.3 Hybrid to 46.8 Fast).
Jargon
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.