@DataChaz: @NVIDIA just dropped LocateAnything, making object detection ~10x faster by fixing one core bottleneck: How the model w…

X AI KOLs Following Models
object-detection nvidia open-source speed-improvement visual-grounding

Summary

NVIDIA released LocateAnything, an open-source model that achieves ~10x faster object detection by predicting all coordinates simultaneously instead of sequentially, reaching 12.7 FPS on a single H100 and outperforming 32B parameter models.

@NVIDIA just dropped LocateAnything, making object detection ~10x faster by fixing one core bottleneck: How the model writes coordinates. Standard AI models do visual grounding the slow way. They predict coordinates piece by piece: Token 1, Token 2, Token 3, Token 4 etc. LocateAnything stops treating boxes like text. It looks at an object and predicts the complete set of coordinates at the exact same time. Because it’s not waiting on itself to finish a "sentence," the speed gains are incredible: → 12.7 frames per second on a single H100 → Dominates across dense object detection and GUI navigation → Outperforms much heavier 32-billion parameter models .. and it's trained on a massive custom dataset of 138 million queries and 785 million bounding boxes. Best part? It's 100% free and open-source. I added the repo link in the ↓
Original Article
View Cached Full Text

Cached at: 06/17/26, 11:52 AM

@NVIDIA just dropped LocateAnything, making object detection ~10x faster by fixing one core bottleneck:

How the model writes coordinates.

Standard AI models do visual grounding the slow way.

They predict coordinates piece by piece: Token 1, Token 2, Token 3, Token 4 etc.

LocateAnything stops treating boxes like text.

It looks at an object and predicts the complete set of coordinates at the exact same time.

Because it’s not waiting on itself to finish a “sentence,” the speed gains are incredible:

→ 12.7 frames per second on a single H100 → Dominates across dense object detection and GUI navigation → Outperforms much heavier 32-billion parameter models

.. and it’s trained on a massive custom dataset of 138 million queries and 785 million bounding boxes.

Best part?

It’s 100% free and open-source.

I added the repo link in the ↓

Similar Articles

@VincentLogic: NVIDIA's newly open-sourced LocateAnything model is really impressive. The previous visual grounding models generated coordinates digit by digit (like squeezing toothpaste), slow and unstable. This new model uses "parallel bounding box decoding" to predict complete coordinates in one step, much faster and more accurate...

X AI KOLs Timeline

NVIDIA has open-sourced the LocateAnything model, using parallel bounding box decoding technology to predict complete coordinates in one step, fast and accurate. The model has only 3B parameters and can run on consumer-grade GPUs, supporting video object localization, UI recognition, OCR, and other tasks.