README.md

---
license: apache-2.0
pipeline_tag: keypoint-detection
---

# Talking Points: Describing and Localizing Pixels

This repository contains the official implementation of the **Talking Points** framework, presented in the paper [Talking Points: Describing and Localizing Pixels](https://huggingface.co/papers/2510.14583).

Vision-language models have achieved remarkable success in cross-modal understanding, but often remain limited to object-level or region-level grounding, lacking the capability for pixel-precise keypoint comprehension through natural language. Talking Points introduces a novel framework for pixel-level grounding, consisting of two complementary components:
-   A **Point Descriptor** that generates rich, contextual descriptions of individual keypoints.
-   A **Point Localizer** that regresses precise pixel coordinates from these descriptions.

Unlike prior work that relies on templated prompts or keypoint names, our approach produces free-form, coarse-to-fine descriptions that situate keypoints within their visual context.

<div align="center">
  <img src="https://github.com/matanr/Talking_Points/raw/master/images/teaser.png" width="100%"/>
</div>

For more detailed information, including installation instructions, dataset creation, training scripts, and evaluation protocols, please refer to the [official GitHub repository](https://github.com/matanr/Talking_Points).

## Acknowledgments
This repository is built upon and incorporates code from [OMG-Seg and OMG-LLaVA](https://github.com/lxtGH/OMG-Seg/tree/main). In addition, it uses the code from [LLaVA](https://github.com/haotian-liu/LLaVA).

## License
This project follows the [Apache-2.0 license](https://github.com/haotian-liu/LLaVA?tab=Apache-2.0-1-ov-file), for the respect of both [LLaVA](https://github.com/haotian-liu/LLaVA) and [XTuner](https://github.com/InternLM/xtuner) codebase.