Bin Xiao
Member of Technical Staff at Microsoft; first author of Florence-2 and former lead of Florence models
Bin Xiao is a Member of Technical Staff on the Microsoft Superintelligence team, working on multimodal foundation models, computer vision and agentic AI. Xiao is the first author of "Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks", presented as an oral paper at CVPR 2024, and Xiao's site says Xiao led the Florence-1 and Florence-2 vision foundation models from 2020 to 2023. Xiao also led the Phi-3 Vision and Phi-3.5 Vision models and co-authored the HRNet and CvT papers.12
Research
Xiao's early work covered visual representation learning and human pose estimation, including "Simple Baselines for Human Pose Estimation and Tracking" (ECCV 2018) as first author, the HRNet paper (CVPR 2019) and CvT, which added convolutions to vision transformers (ICCV 2021). Xiao's site lists first place in the single-person pose estimation track of the 2019 Look into Person challenge.1
From 2020 to 2023 Xiao led the Florence-1 and Florence-2 vision foundation models; the Florence-2 paper, posted in November 2023 with Lu Yuan as senior author, unifies captioning, detection, grounding and segmentation behind text prompts. In 2024 Xiao led Phi-3 Vision and Phi-3.5 Vision.12
Recent work
Xiao's site describes work on multimodal Llama post-training for visual understanding in 2024 and 2025, and since 2025 on post-training for reasoning, coding and agentic models, including Microsoft AI's MAI-Thinking-1 and MAI-Code-1-Flash releases in 2026.1
Work
Microsoft Florence-2Open weights
First author of the Florence-2 paper, which presents a prompt-based sequence-to-sequence vision model for captioning, object detection, grounding and segmentation.21
Florence-2-largeOpen weights
Led the Florence-2 vision foundation model work from which the Florence-2-large checkpoint was released.12
Phi-3 Vision and Phi-3.5 Vision (external site: leoxiaobin.github.io)Open weights
Led Microsoft's Phi-3 Vision and Phi-3.5 Vision compact multimodal language models in 2024.1
Career timeline
2018
Published "Simple Baselines for Human Pose Estimation and Tracking" at ECCV 2018 as first author.1
2019
Co-authored the HRNet paper on high-resolution representation learning at CVPR 2019.1
2020
Began leading the Florence vision foundation model work.1
Nov 2023
Posted the Florence-2 paper as first author.2
2024
Led the Phi-3 Vision and Phi-3.5 Vision multimodal models; Florence-2 presented as a CVPR 2024 oral.1
2026
Contributed to Microsoft AI's MAI-Thinking-1 and MAI-Code-1-Flash releases.1
Honors and recognition
- First place, Look into Person Challenge (single-person pose estimation) · Look into Person (LIP) Challenge, 20191
Related people
More sourced profiles at the same organization, in the same field, or from the same school.
- Adam FourneyAlso at Microsoft Research
- Aditya RameshMultimodal models
- Andreas C. MüllerAlso at Microsoft
- Alexander KirillovMultimodal models
- Charles LamannaAlso at Microsoft
- Ali FarhadiComputer vision
- Christopher BishopAlso at Microsoft Research
- Christoph FeichtenhoferComputer vision
- Eric HorvitzAlso at Microsoft
- Christopher ClarkMultimodal models
- Furu WeiAlso at Microsoft Research
- Cordelia SchmidComputer vision
Is this you?
Profiles cover public professional work only, and every line cites a source. If something is out of date, missing or wrong, send a correction with a link to an official page and we will update it. Suggest a correction
Add this badge to your website or GitHub profile README. It links back to this page. It is not an endorsement or a rating.
Markdown (for a README)
[](https://unitedstatesofamericasuperintelligence.com/people/bin-xiao/)
HTML
<a href="https://unitedstatesofamericasuperintelligence.com/people/bin-xiao/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/profile.svg" alt="Bin Xiao — Profile on USASI" height="20"></a>
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.