Min Xu

Assistant Professor at Carnegie Mellon University

[Min Xu @xumin100]

#CVPR2025 Thrilled to share that our paper "DiffCAM: Data-Driven Saliency Maps by Capturing Feature Differences" has been accepted at CVPR 2025!

In this work, we present DiffCAM, a data-driven XAI (explainable AI) method that generates saliency maps by identifying feature differences from reference data distributions, without relying on predictions and gradients. Unlike mainstream approaches, DiffCAM adopts a data-centric perspective, offering meaningful explanations applicable to a wide range of models, including supervised/self-supervised learning models and CNN/ViT architectures.

Stay tuned for Arxiv and GitHub links!

Huge thanks to my amazing team members: Xingjian Li, Qiming Zhao, Neelesh Bisht, @duran_rafid , Jin Yu Kim and Bryan Zhang. Excited to present this at #CVPR2025 #ExplainableAI #SaliencyMaps #XAI #Interpretability #CVPR

[Jiayuan Zhu @JiayuanZhu_]

📣 CFP – AIMedHealth Bridge Program @ Association for the Advancement of Artificial Intelligence (@RealAAAI) 2026, Singapore!

We invite submissions to the 2nd AAAI Bridge Program on AI for Medicine and Healthcare, taking place in Singapore, January 20–21, 2026.

This Bridge addresses the critical gap between the rapid progress of AI in medical research and its limited integration into real-world clinical practice. We welcome contributions from both the AI and clinical communities to foster meaningful collaboration and innovation.

📝 Submission deadline: October 31, 2025
📍 Submit here: https://t.co/itWk1riM7t
🌐 More details: https://t.co/YIJoFaz4xM

Kudos to our amazing organizing team @JundeMorsenWu , @PeterPanJZ , Fenglin Liu, Luyang Luo, @xumin100, @JinYueming, David Clifton, @pranavrajpurkar, @DanielRueckert 🙌

#AAAI #MedicalAI #AIResearch #MedTech #DeepLearning #MedicalResearch

[Min Xu @xumin100]

Excited to share that we have four papers at CVPR 2025!  
If you’ll be in Nashville, don’t miss the chance to stop by and chat with us!

🟡 DiffCAM: Data-Driven Saliency Maps by Capturing Feature Differences (**Highlight**)  
📍 Poster Session 2 · ExHall D · Poster #472  
📅 June 13 (Fri) · 5:00–7:00 PM

🟡 BOE-ViT: Boosting Orientation Estimation with Equivariance in Self-Supervised 3D Subtomogram Alignment  
📍 Poster Session 6 · ExHall D · Poster #306  
📅 June 15 (Sun) · 5:00–7:00 PM

🟡 Multimodal Generalized Category Discovery (MM-GCD)  
📍 TMM-OpenWorld 2025 · Room 104E  
📅 June 11 (Wed) · 4:30–5:30 PM

🟡 Multimodal Foundation Model for Protein Retrieval  
📍 MMFM-BIOMED Workshop · Room 107A  
📅 June 11 (Wed) · 3:00–3:40 PM

Big thanks to all my co-authors — couldn’t have done it without you!

#CVPR2025 #ComputerVision #XAI #VisionTransformers #SubtomogramAlignment #MultimodalLearning #ProteinRetrieval

[Min Xu @xumin100]

#CVPR2025 Happy to share that our paper “BOE-ViT: Boosting Orientation Estimation with Equivariance in Self-Supervised 3D Subtomogram Alignment” has been accepted at CVPR 2025!
In this work, we introduce the first Vision Transformer for cryo-ET subtomogram alignment, addressing two critical challenges:
Limited labels & high noise: Current methods struggle with severe noise and scarce positional labels in 3D cryoET subtomograms.
Orientation estimation limitations: Existing vision models have difficulty accurately estimating orientations crucial for alignment.
BOE-ViT explores advanced network architectures when high-quality real data for large-scale pre-training is unavailable. Through equivariant design modifications, we enhance both rotation and translation estimation without requiring explicit annotations. Experiments demonstrate superior performance across various datasets and noise conditions.

Arxiv and GitHub links are coming soon!

Huge thanks to Runmin Jiang and Xingjian Li, who are leading the project, and to the amazing team members: Jackson Daggett, Shriya Pingulkar, Yizhou Zhao, Priyanshu Jha Dhingra, Daniel Brown, Qifeng Wu, and Xiangrui Zeng. #CVPR #CVPR2025 #cryoET #Equivariance #ViT #subtomogram-alignment

[Min Xu @xumin100]

MfH uses generatively painted humans as landmarks to bring relative depth to the metric space, handling in-the-wild scenes with remarkable generalization.Curious about the details? Let's explore more here:
📷 Full paper: https://t.co/zeKdB0qPZR
📷 Code: https://t.co/vmFD905rDp

[Min Xu @xumin100]

Ready to reimagine monocular metric depth estimation? Welcome to stop by Poster Session 5, East at #NeurIPS2024, and explore my PhD student @YizhouZhao42 's Metric from Human (MfH). No training with metric depth annotations is required!

[MICCAI Society @MICCAI_Society]

📢 We are thrilled to announce that #MICCAI2026 will be held in Abu Dhabi, UAE! 👏 This will be our first #MICCAI conference in the Middle East.
Save the date!
🗓️  Oct 4-8, 2026
🔗https://t.co/5jiVnxosaL
@MiccaiStudents @RMiccai @WomenInMICCAI
#ai #deeplearning #imaging

[JUNDE WU @JundeMorsenWu]

I’m thrilled to announce that our work, MedSegDiff-V2, has been selected as AAAI 2024’s Most Influential Paper! 🥳
https://t.co/aAkMAZE0i3
In this paper, we build on MedSegDiff-V1 by integrating transformers into the diffusion model for medical image segmentation, with exciting insights on scaling laws! (Details will be covered in our upcoming extended journal version.)

Read our paper: https://t.co/69GeWQMb4g
Access the code: https://t.co/pVhtcqZeHM

A huge thank you to Wei Ji, @xumin100, @Hz_MedAI, @JinYueming, for the invaluable collaboration!

[JUNDE WU @JundeMorsenWu]

I’m excited to announce that our 1st AI for Medicine and Healthcare (AI4MedHealth) bridge program, will be held with AAAI @RealAAAI in Philadelphia on February 25-26, 2025. Please consider submitting this year. Our freshly baked website is here😁: https://t.co/lfj5GDIm0K

For those unfamiliar with the AAAI Bridge Program, think of it as a special workshop that emphasizes the interdisciplinary collaboration between AI and other communities, directed towards a common goal.

Thrilled to be organizing this one-of-a-kind program connecting AI and Medicine alongside my amazing colleagues! @JinYueming @xumin100 @JiayuanZhu_ @bwpapiez @sarim_ather, and Prof. Alison Nobel, Dr. Alex Novak

[Jiayuan Zhu @JiayuanZhu_]

We are excited to share our 1st AAAI @RealAAAI Bridge Program - AI for Medicine and Healthcare (AI4MedHealth). This event will take place in Philadelphia on February 25-26, 2025. Feel free to check out our newborn website for details: https://t.co/YIJoFazCnk

We are calling for submissions which integrate AI into medical settings. Thank you for my amazing co-organisers @JundeMorsenWu @xumin100 @JinYueming @bwpapiez @sarim_ather, and Dr. Alex Novak, Prof. Alison Noble 🎉

#AAAI #AIResearch #MedTech #MedicalResearch

[Min Xu @xumin100]

We're pleased to introduce CryoSAM, our new approach to Cryo-ET segmentation. This training-free framework leverages foundation models for efficient segmentation, reducing the need for manual annotations. Accepted by #MICCAI
https://t.co/6YjcGrQQio

@cryoem_papers #cryoem_papers

[Yashika Jain @yashikajain201]

Our #ECCV2024 paper, "Cross-Domain Learning for Video Anomaly Detection with Limited Supervision", demonstrates the effectiveness of uncertainty-aware integration of external, unlabeled data with weakly-labeled source data to enhance the cross-domain generalization of VAD models.

[Min Xu @xumin100]

We're pleased to introduce CryoSAM, our new approach to Cryo-ET segmentation. This training-free framework leverages foundation models for efficient segmentation, reducing the need for manual annotations. Accepted by #MICCAI
https://t.co/6YjcGrQQio

@cryoem_papers #cryoem_papers

[Yashika Jain @yashikajain201]

Our paper "Cross-Domain Learning for Video Anomaly Detection with Limited Supervision" got accepted at #ECCV2024!

My first-ever paper :)

Super grateful to @alidabouei and @xumin100 for the guidance and opportunity!

[Min Xu @xumin100]

Welcome to check out our recent CVPR paper on medical image segmentation!

[JUNDE WU @JundeMorsenWu]

I'm excited to announce that we will be presenting our #CVPR paper, "One-prompt to Segment All Medical Images," at Arch 4A-E from 10:30 to 12:00 on June 20th. Come if you're interested🥳

Prof. @xumin100 will be there in person, and I will be online, as I'm unfortunately still unable to obtain my US visa.

I'd also like to give a special shout-out to my amazing co-authors, @anna_jiaaaa and @JinYueming , for their substantial contributions. Could not have a chance to make it without them🎉🎉

https://t.co/BBl5vErf4U

[Min Xu @xumin100]

Manuscript of our recent research available: Enhancing Weakly Supervised 3D Medical Image Segmentation through Probabilistic-aware Learning. Welcome to check out.

https://t.co/RcbhD9wtD5

[Min Xu @xumin100]

Welcome to take a look at the manuscript detailing our recent research in cryo-ET denoising and simulation.
https://t.co/7OhduUy3xb

[Min Xu @xumin100]

And check out the paper here! https://t.co/6NTdeHPpJZ

[Min Xu @xumin100]

Excited to share our latest research at #AAAI24! Our paper introduces a general active learning algorithm using noise stability. Watch the presentation video for insights: https://t.co/e1AAPryngV

[Min Xu @xumin100]

Welcome to check out our latest work at #AAAI2024: MedSegDiff-v2: Diffusion based Medical Image Segmentation with Transformers

[JUNDE WU @JundeMorsenWu]

Excited to share our latest work: MedSegDiff-v2: Diffusion based Medical Image Segmentation with Transformers, posted at #AAAI24 conference🥳. Check out our paper📄 here: https://t.co/ie5uGau8En and project🔨 here: https://t.co/L7CRzmnfN2. Hope you find it interesting! 👾