Computational Imaging Group (CIG)
@uwcig
Followers
1K
Following
111
Media
135
Statuses
477
Computational Imaging Group (CIG) at UW-Madison. Computational Imaging, Biomedical Imaging, Optimization, and AI. Director @ukmlv.
Madison, WI
Joined November 2017
Overview of the Computational Imaging Group (CIG) at WashU by @ukmlv, @s_shirin_s, and @YuyangHu_666. https://t.co/hBn5oHjMqk
0
1
4
Introducing ShaRP for Image Reconstruction/Restoration. Thanks to @YuyangHu_666 for sharing this illustration. 🔗 https://t.co/4QvQKeVrYy 🔗 https://t.co/nvLW5IVlsb
0
1
15
CIG is in the process of moving from WashU to UW–Madison. We will be updating and sharing our new website soon. Meanwhile, we will be in touch here.
0
0
2
We hope this makes diffusion models easier to understand and extend, especially for the signal processing community. 📄 https://t.co/MH8BbtAaqm 🔗
github.com
Contribute to wustl-cig/randomwalk_diffusion development by creating an account on GitHub.
0
0
1
We derive training and sampling templates that cover NCSN, DDPM, and SGM-SDE as special cases. No need for separate proofs—only different parameter choices (noise schedule, step-size, temperature).
1
0
0
Once you have the score, sampling is just a random walk in a potential field: xₖ₊₁ = xₖ + step·∇ log pₖ(xₖ) + noise. This connects diffusion models to classical stochastic differential equations.
1
0
0
At the core lies Tweedie’s formula—a textbook link between the MMSE denoiser and the score: ∇ log p(z) = (E[X|z] − z)/σ². That’s it. Learn to denoise, and you already know the score!
1
0
0
Diffusion models power today’s AI image generators. But theory is scattered: different presentation for NCSN, DDPM, SGM-SDE. We present them using the samesimple principle—a sequence of random walks driven by the score (∇ log p(x)).
1
0
0
New in IEEE Signal Processing Magazine: “Random Walks with Tweedie” - our tutorial on modern diffusion models (NCSN, DDPM, SGM-SDE). We show that they’re all random walks guided by the Tweedie’s formula.
1
2
5
New paper on Out-of-Distribution (OOD) detection. OOD is a critical challenge for deploying AI/ML models safely, as models can become highly confident when encountering data outside their training set. EigenScore is our new OOD detection method built on diffusion models 👇
0
1
6
🎉 "Kernel Density Steering: Inference-Time Scaling via Mode Seeking for Image Restoration" was accepted to NeurIPS 2025. This paper is part of @YuyangHu_666's internship at Google with @2ptmvd and @docmilanfar. 👉 Read here: https://t.co/yOBAeWdXJD.
arxiv.org
Diffusion models show promise for image restoration, but existing methods often struggle with inconsistent fidelity and undesirable artifacts. To address this, we introduce Kernel Density Steering...
0
2
7
New paper "Analysis Plug-and-Play Methods for Imaging Inverse Problems" considers an alternative PnP formulation in which the prior is imposed on a transformed representation of the image, such as its gradient. 👉 Read here:
arxiv.org
Plug-and-Play Priors (PnP) is a popular framework for solving imaging inverse problems by integrating learned priors in the form of denoisers trained to remove Gaussian noise from images. In...
0
1
7
Thrilled to share that my paper from my internship at Mitsubishi Electric Research Laboratories (@merl_news) last summer, “Multimodal Diffusion Bridge with Attention-Based SAR Fusion for Satellite Image Cloud Removal”, has been accepted to IEEE Transactions on Geoscience and
1
4
13
A new study from MIR researchers and collaborators presents a PET/CT quantification pipeline that tracks multiple myeloma progression in mice, overcoming manual analysis limits and enhancing the team's evaluation of treatment response. 🔗: https://t.co/3HYbGuAVfL
#MIRresearch
0
2
8
Great work by @FloraSun9101 and the rest of our big collaborative team. This project was done in collaboration with Prof. Monica Shokeen from @WashUMedMIR.
0
0
1
🚨 New paper accepted in EJNMMI Research! We present a robust PET/CT quantification pipeline for tracking multiple myeloma progression in mice. 🧠 + 📊 + 🐭 = better preclinical insights. Read more: https://t.co/C9E08g5GP1
2
0
7
Headed to ICML in Vancouver (July 12-17)! 🇨🇦 Catch me and @albert_peng_ on Tuesday, July 15, from 4:30 PM - 7:00 PM as we present our poster: "Stochastic Deep Restoration Priors for Imaging Inverse Problems." Let's chat about diffusion models, inverse problems, image/video
🎉Thrilled to share our latest work “Stochastic Deep Restoration Priors for Imaging Inverse Problems”! Big thanks to my amazing collaborators @ukmlv @2ptmvd @docmilanfar @WeijieGan1 @albert_peng_. ⭑ Preprint: https://t.co/fuDI4akCsa. ⭑ Website: https://t.co/ZE9quTOim2.
0
3
13
The CIG team Summer 2025. We are missing Chicago and Eddie who are doing internships.
1
1
12
New paper "Kernel Density Steering: Inference-Time Scaling via Mode Seeking for Image Restoration." This paper is part of @YuyangHu_666's internship at @GoogleAI focused on generating high-fidelity results using diffusion modes for image restoration. 🔗
arxiv.org
Diffusion models show promise for image restoration, but existing methods often struggle with inconsistent fidelity and undesirable artifacts. To address this, we introduce Kernel Density Steering...
1
1
6
"Random Walks with Tweedie: A Unified View of Score-Based Diffusion Models" was accepted to IEEE Signals Processing Magazine (@IEEEspm). It presents a self-contained discussion of score-based diffusion models focused at the signal processing community. 🔗
1
1
8