Junyu Chen

Se-In Jang, Tinsu Pan, Ye Li et al.

Position emission tomography (PET) is widely used in clinics and research due to its quantitative merits and high sensitivity, but suffers from low signal-to-noise ratio (SNR). Recently convolutional neural networks (CNNs) have been widely used to improve PET image quality. Though successful and efficient in local feature extraction, CNN cannot capture long-range dependencies well due to its limited receptive field. Global multi-head self-attention (MSA) is a popular approach to capture long-range information. However, the calculation of global MSA for 3D images has high computational costs. In this work, we proposed an efficient spatial and channel-wise encoder-decoder transformer, Spach Transformer, that can leverage spatial and channel information based on local and global MSAs. Experiments based on datasets of different PET tracers, i.e., $^{18}$F-FDG, $^{18}$F-ACBC, $^{18}$F-DCFPyL, and $^{68}$Ga-DOTATATE, were conducted to evaluate the proposed framework. Quantitative results show that the proposed Spach Transformer framework outperforms state-of-the-art deep learning architectures. Our codes are available at https://github.com/sijang/SpachTransformer

Aditi, Niket Agarwal, Arslan Ali et al.

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI -- effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation's OpenMDW-1.1 https://openmdw.ai/license/1-1/ License at https://github.com/nvidia/cosmos}{github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3 . The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos3 .

Yihao Liu, Junyu Chen, Lianrui Zuo et al.

Deformable image registration establishes non-linear spatial correspondences between fixed and moving images. Deep learning-based deformable registration methods have been widely studied in recent years due to their speed advantage over traditional algorithms as well as their better accuracy. Most existing deep learning-based methods require neural networks to encode location information in their feature maps and predict displacement or deformation fields though convolutional or fully connected layers from these high-dimensional feature maps. In this work, we present Vector Field Attention (VFA), a novel framework that enhances the efficiency of the existing network design by enabling direct retrieval of location correspondences. VFA uses neural networks to extract multi-resolution feature maps from the fixed and moving images and then retrieves pixel-level correspondences based on feature similarity. The retrieval is achieved with a novel attention module without the need of learnable parameters. VFA is trained end-to-end in either a supervised or unsupervised manner. We evaluated VFA for intra- and inter-modality registration and for unsupervised and semi-supervised registration using public datasets, and we also evaluated it on the Learn2Reg challenge. Experimental results demonstrate the superior performance of VFA compared to existing methods. The source code of VFA is publicly available at https://github.com/yihao6/vfa/.

Junyu Chen, Yihao Liu, Shuwen Wei et al.

Deep learning technologies have dramatically reshaped the field of medical image registration over the past decade. The initial developments, such as regression-based and U-Net-based networks, established the foundation for deep learning in image registration. Subsequent progress has been made in various aspects of deep learning-based registration, including similarity measures, deformation regularizations, network architectures, and uncertainty estimation. These advancements have not only enriched the field of image registration but have also facilitated its application in a wide range of tasks, including atlas construction, multi-atlas segmentation, motion estimation, and 2D-3D registration. In this paper, we present a comprehensive overview of the most recent advancements in deep learning-based image registration. We begin with a concise introduction to the core concepts of deep learning-based image registration. Then, we delve into innovative network architectures, loss functions specific to registration, and methods for estimating registration uncertainty. Additionally, this paper explores appropriate evaluation metrics for assessing the performance of deep learning models in registration tasks. Finally, we highlight the practical applications of these novel techniques in medical imaging and discuss the future prospects of deep learning-based image registration.

Yecheng Wu, Zhuoyang Zhang, Junyu Chen et al.

VILA-U is a Unified foundation model that integrates Video, Image, Language understanding and generation. Traditional visual language models (VLMs) use separate modules for understanding and generating visual content, which can lead to misalignment and increased complexity. In contrast, VILA-U employs a single autoregressive next-token prediction framework for both tasks, eliminating the need for additional components like diffusion models. This approach not only simplifies the model but also achieves near state-of-the-art performance in visual language understanding and generation. The success of VILA-U is attributed to two main factors: the unified vision tower that aligns discrete visual tokens with textual inputs during pretraining, which enhances visual perception, and autoregressive image generation can achieve similar quality as diffusion models with high-quality dataset. This allows VILA-U to perform comparably to more complex models using a fully token-based autoregressive framework.

Gary Y. Li, Junyu Chen, Se-In Jang et al.

Radiotherapy (RT) combined with cetuximab is the standard treatment for patients with inoperable head and neck cancers. Segmentation of head and neck (H&N) tumors is a prerequisite for radiotherapy planning but a time-consuming process. In recent years, deep convolutional neural networks have become the de facto standard for automated image segmentation. However, due to the expensive computational cost associated with enlarging the field of view in DCNNs, their ability to model long-range dependency is still limited, and this can result in sub-optimal segmentation performance for objects with background context spanning over long distances. On the other hand, Transformer models have demonstrated excellent capabilities in capturing such long-range information in several semantic segmentation tasks performed on medical images. Inspired by the recent success of Vision Transformers and advances in multi-modal image analysis, we propose a novel segmentation model, debuted, Cross-Modal Swin Transformer (SwinCross), with cross-modal attention (CMA) module to incorporate cross-modal feature extraction at multiple resolutions.To validate the effectiveness of the proposed method, we performed experiments on the HECKTOR 2021 challenge dataset and compared it with the nnU-Net (the backbone of the top-5 methods in HECKTOR 2021) and other state-of-the-art transformer-based methods such as UNETR, and Swin UNETR. The proposed method is experimentally shown to outperform these comparing methods thanks to the ability of the CMA module to capture better inter-modality complimentary feature representations between PET and CT, for the task of head-and-neck tumor segmentation.

Jun Li, Junyu Chen, Yucheng Tang et al.

Transformer, the latest technological advance of deep learning, has gained prevalence in natural language processing or computer vision. Since medical imaging bear some resemblance to computer vision, it is natural to inquire about the status quo of Transformers in medical imaging and ask the question: can the Transformer models transform medical imaging? In this paper, we attempt to make a response to the inquiry. After a brief introduction of the fundamentals of Transformers, especially in comparison with convolutional neural networks (CNNs), and highlighting key defining properties that characterize the Transformers, we offer a comprehensive review of the state-of-the-art Transformer-based approaches for medical imaging and exhibit current research progresses made in the areas of medical image segmentation, recognition, detection, registration, reconstruction, enhancement, etc. In particular, what distinguishes our review lies in its organization based on the Transformer's key defining properties, which are mostly derived from comparing the Transformer and CNN, and its type of architecture, which specifies the manner in which the Transformer and CNN are combined, all helping the readers to best understand the rationale behind the reviewed approaches. We conclude with discussions of future perspectives.

Zhangxing Bian, Shuwen Wei, Yihao Liu et al.

Tagged magnetic resonance imaging (tMRI) has been employed for decades to measure the motion of tissue undergoing deformation. However, registration-based motion estimation from tMRI is difficult due to the periodic patterns in these images, particularly when the motion is large. With a larger motion the registration approach gets trapped in a local optima, leading to motion estimation errors. We introduce a novel "momenta, shooting, and correction" framework for Lagrangian motion estimation in the presence of repetitive patterns and large motion. This framework, grounded in Lie algebra and Lie group principles, accumulates momenta in the tangent vector space and employs exponential mapping in the diffeomorphic space for rapid approximation towards true optima, circumventing local optima. A subsequent correction step ensures convergence to true optima. The results on a 2D synthetic dataset and a real 3D tMRI dataset demonstrate our method's efficiency in estimating accurate, dense, and diffeomorphic 2D/3D motion fields amidst large motion and repetitive patterns.

Md Yousuf Harun, Jhair Gallardo, Junyu Chen et al.

Continual learning (CL) in deep neural networks (DNNs) involves incrementally accumulating knowledge in a DNN from a growing data stream. A major challenge in CL is that non-stationary data streams cause catastrophic forgetting of previously learned abilities. A popular solution is rehearsal: storing past observations in a buffer and then sampling the buffer to update the DNN. Uniform sampling in a class-balanced manner is highly effective, and better sample selection policies have been elusive. Here, we propose a new sample selection policy called GRASP that selects the most prototypical (easy) samples first and then gradually selects less prototypical (harder) examples. GRASP has little additional compute or memory overhead compared to uniform selection, enabling it to scale to large datasets. Compared to 17 other rehearsal policies, GRASP achieves higher accuracy in CL experiments on ImageNet. Compared to uniform balanced sampling, GRASP achieves the same performance with 40% fewer updates. We also show that GRASP is effective for CL on five text classification datasets.

Ye Li, Jianan Cui, Junyu Chen et al.

In PET, the amount of relative (signal-dependent) noise present in different body regions can be significantly different and is inherently related to the number of counts present in that region. The number of counts in a region depends, in principle and among other factors, on the total administered activity, scanner sensitivity, image acquisition duration, radiopharmaceutical tracer uptake in the region, and patient local body morphometry surrounding the region. In theory, less amount of denoising operations is needed to denoise a high-count (low relative noise) image than images a low-count (high relative noise) image, and vice versa. The current deep-learning-based methods for PET image denoising are predominantly trained on image appearance only and have no special treatment for images of different noise levels. Our hypothesis is that by explicitly providing the local relative noise level of the input image to a deep convolutional neural network (DCNN), the DCNN can outperform itself trained on image appearance only. To this end, we propose a noise-level-aware framework denoising framework that allows embedding of local noise level into a DCNN. The proposed is trained and tested on 30 and 15 patient PET images acquired on a GE Discovery MI PET/CT system. Our experiments showed that the increases in both PSNR and SSIM from our backbone network with relative noise level embedding (NLE) versus the same network without NLE were statistically significant with p<0.001, and the proposed method significantly outperformed a strong baseline method by a large margin.

Junyu Chen, Yihao Liu, Yufan He et al.

Transformers have recently shown promise for medical image applications, leading to an increasing interest in developing such models for medical image registration. Recent advancements in designing registration Transformers have focused on using cross-attention (CA) to enable a more precise understanding of spatial correspondences between moving and fixed images. Here, we propose a novel CA mechanism that computes windowed attention using deformable windows. In contrast to existing CA mechanisms that require intensive computational complexity by either computing CA globally or locally with a fixed and expanded search window, the proposed deformable CA can selectively sample a diverse set of features over a large search window while maintaining low computational complexity. The proposed model was extensively evaluated on multi-modal, mono-modal, and atlas-to-patient registration tasks, demonstrating promising performance against state-of-the-art methods and indicating its effectiveness for medical image registration. The source code for this work will be available after publication.

Junyu Chen, Jungang Li, Jing Xiong et al.

Large language model (LLM) inference is often bounded by memory footprint and memory bandwidth in resource-constrained deployments, making quantization a fundamental technique for efficient serving. While post-training quantization (PTQ) maintains high fidelity at 4-bit, it deteriorates at 2-3 bits. Fundamentally, existing methods enforce a shape-invariant quantization grid (e.g., the fixed uniform intervals of UINT2) for each group, severely restricting the feasible set for error minimization. To address this, we propose Bit-Plane Decomposition Quantization (BPDQ), which constructs a variable quantization grid via bit-planes and scalar coefficients, and iteratively refines them using approximate second-order information while progressively compensating quantization errors to minimize output discrepancy. In the 2-bit regime, BPDQ enables serving Qwen2.5-72B on a single RTX 3090 with 83.85% GSM8K accuracy (vs. 90.83% at 16-bit). Moreover, we provide theoretical analysis showing that the variable grid expands the feasible set, and that the quantization process consistently aligns with the optimization objective in Hessian-induced geometry. Code: github.com/KingdalfGoodman/BPDQ.

Ye Li, Junyu Chen, Se-in Jang et al.

Inspired by the recent success of Transformers for Natural Language Processing and vision Transformer for Computer Vision, many researchers in the medical imaging community have flocked to Transformer-based networks for various main stream medical tasks such as classification, segmentation, and estimation. In this study, we analyze, two recently published Transformer-based network architectures for the task of multimodal head-and-tumor segmentation and compare their performance to the de facto standard 3D segmentation network - the nnU-Net. Our results showed that modeling long-range dependencies may be helpful in cases where large structures are present and/or large field of view is needed. However, for small structures such as head-and-neck tumor, the convolution-based U-Net architecture seemed to perform well, especially when training dataset is small and computational resource is limited.

Junyu Chen, Yihao Liu, Yufan He et al.

In the past, optimization-based registration models have used spatially-varying regularization to account for deformation variations in different image regions. However, deep learning-based registration models have mostly relied on spatially-invariant regularization. Here, we introduce an end-to-end framework that uses neural networks to learn a spatially-varying deformation regularizer directly from data. The hyperparameter of the proposed regularizer is conditioned into the network, enabling easy tuning of the regularization strength. The proposed method is built upon a Transformer-based model, but it can be readily adapted to any network architecture. We thoroughly evaluated the proposed approach using publicly available datasets and observed a significant performance improvement while maintaining smooth deformation. The source code of this work will be made available after publication.

Dongyun Zou, Zhuoyang Zhang, Junyu Chen et al.

We introduce JetViT, a novel family of hybrid-architecture Vision Transformer (ViT) models that match the accuracy of state-of-the-art full-attention vision foundation models while achieving substantially higher inference efficiency on high-resolution images. At the core of our approach is Post-Training Attention Search, a post-training acceleration framework that converts pre-trained full-attention ViTs into efficient hybrid-attention variants by identifying and replacing redundant full-attention blocks with linear or window-attention blocks. By inheriting the MLP and attention weights from the base model, Post-Training Attention Search efficiently explores the architectural design space through three key steps: (1) optimizing the linear-attention block design; (2) finding the best combination of linear-attention and window-attention blocks; and (3) identifying and preserving critical full-attention blocks. We evaluate JetViT on two representative high-resolution vision foundation models, DINOv3 and DepthAnythingV2. On the NVIDIA H100 GPU, JetViT achieves up to 1.79x higher throughput and up to 44.81% lower latency without sacrificing accuracy. We will release our code and accelerated ViT models soon.

Zhangxing Bian, Shuwen Wei, Samuel W. Remedios et al.

Tagged MRI enables tracking internal tissue motion non-invasively. It encodes motion by modulating anatomy with periodic tags, which deform along with tissue. However, the entanglement between anatomy, tags and motion poses significant challenges for post-processing. The existence of tags and imaging blur hinders downstream tasks such as segmenting anatomy. Tag fading, due to T1-relaxation, disrupts the brightness constancy assumption for motion tracking. For decades, these challenges have been handled in isolation and sub-optimally. In contrast, we introduce a blind and nonlinear inverse framework for tagged MRI that, for the first time, unifies these tasks: anatomical image recovery, high-resolution cine image synthesis, and motion estimation. At its core, the synergy of MR physics and generative priors enables us to blindly estimate the unknown forward imaging models and high-resolution underlying anatomy, while simultaneously tracking 3D diffeomorphic Lagrangian motion over time. Experiments on tagged brain MRI demonstrate that our approach yields high-resolution anatomy images, cine images, and more accurate motion than specialized methods.

Qing Guo, Xinhang Li, Junyu Chen et al.

Leveraging large language models (LLMs) in traffic signal control (TSC) improves optimization efficiency and interpretability compared to traditional reinforcement learning (RL) methods. However, existing LLM-based approaches are limited by fixed time signal durations and are prone to hallucination errors, while RL methods lack robustness in signal timing decisions and suffer from poor generalization. To address these challenges, this paper proposes HeraldLight, a dual LLMs architecture enhanced by Herald guided prompts. The Herald Module extracts contextual information and forecasts queue lengths for each traffic phase based on real-time conditions. The first LLM, LLM-Agent, uses these forecasts to make fine grained traffic signal control, while the second LLM, LLM-Critic, refines LLM-Agent's outputs, correcting errors and hallucinations. These refined outputs are used for score-based fine-tuning to improve accuracy and robustness. Simulation experiments using CityFlow on real world datasets covering 224 intersections in Jinan (12), Hangzhou (16), and New York (196) demonstrate that HeraldLight outperforms state of the art baselines, achieving a 20.03% reduction in average travel time across all scenarios and a 10.74% reduction in average queue length on the Jinan and Hangzhou scenarios. The source code is available on GitHub: https://github.com/BUPT-ANTlab/HeraldLight.

Junyu Chen, Jie An, Hanjia Lyu et al.

Visual-textual sentiment analysis aims to predict sentiment with the input of a pair of image and text, which poses a challenge in learning effective features for diverse input images. To address this, we propose a holistic method that achieves robust visual-textual sentiment analysis by exploiting a rich set of powerful pre-trained visual and textual prior models. The proposed method consists of four parts: (1) a visual-textual branch to learn features directly from data for sentiment analysis, (2) a visual expert branch with a set of pre-trained "expert" encoders to extract selected semantic visual features, (3) a CLIP branch to implicitly model visual-textual correspondence, and (4) a multimodal feature fusion network based on BERT to fuse multimodal features and make sentiment predictions. Extensive experiments on three datasets show that our method produces better visual-textual sentiment analysis performance than existing methods.

Junyu Chen, Han Cai, Junsong Chen et al.

We present Deep Compression Autoencoder (DC-AE), a new family of autoencoder models for accelerating high-resolution diffusion models. Existing autoencoder models have demonstrated impressive results at a moderate spatial compression ratio (e.g., 8x), but fail to maintain satisfactory reconstruction accuracy for high spatial compression ratios (e.g., 64x). We address this challenge by introducing two key techniques: (1) Residual Autoencoding, where we design our models to learn residuals based on the space-to-channel transformed features to alleviate the optimization difficulty of high spatial-compression autoencoders; (2) Decoupled High-Resolution Adaptation, an efficient decoupled three-phases training strategy for mitigating the generalization penalty of high spatial-compression autoencoders. With these designs, we improve the autoencoder's spatial compression ratio up to 128 while maintaining the reconstruction quality. Applying our DC-AE to latent diffusion models, we achieve significant speedup without accuracy drop. For example, on ImageNet 512x512, our DC-AE provides 19.1x inference speedup and 17.9x training speedup on H100 GPU for UViT-H while achieving a better FID, compared with the widely used SD-VAE-f8 autoencoder. Our code is available at https://github.com/mit-han-lab/efficientvit.

Haotian Tang, Yecheng Wu, Shang Yang et al.

We introduce Hybrid Autoregressive Transformer (HART), an autoregressive (AR) visual generation model capable of directly generating 1024x1024 images, rivaling diffusion models in image generation quality. Existing AR models face limitations due to the poor image reconstruction quality of their discrete tokenizers and the prohibitive training costs associated with generating 1024px images. To address these challenges, we present the hybrid tokenizer, which decomposes the continuous latents from the autoencoder into two components: discrete tokens representing the big picture and continuous tokens representing the residual components that cannot be represented by the discrete tokens. The discrete component is modeled by a scalable-resolution discrete AR model, while the continuous component is learned with a lightweight residual diffusion module with only 37M parameters. Compared with the discrete-only VAR tokenizer, our hybrid approach improves reconstruction FID from 2.11 to 0.30 on MJHQ-30K, leading to a 31% generation FID improvement from 7.85 to 5.38. HART also outperforms state-of-the-art diffusion models in both FID and CLIP score, with 4.5-7.7x higher throughput and 6.9-13.4x lower MACs. Our code is open sourced at https://github.com/mit-han-lab/hart.

Enze Xie, Junsong Chen, Yuyang Zhao et al.

This paper presents SANA-1.5, a linear Diffusion Transformer for efficient scaling in text-to-image generation. Building upon SANA-1.0, we introduce three key innovations: (1) Efficient Training Scaling: A depth-growth paradigm that enables scaling from 1.6B to 4.8B parameters with significantly reduced computational resources, combined with a memory-efficient 8-bit optimizer. (2) Model Depth Pruning: A block importance analysis technique for efficient model compression to arbitrary sizes with minimal quality loss. (3) Inference-time Scaling: A repeated sampling strategy that trades computation for model capacity, enabling smaller models to match larger model quality at inference time. Through these strategies, SANA-1.5 achieves a text-image alignment score of 0.81 on GenEval, which can be further improved to 0.96 through inference scaling with VILA-Judge, establishing a new SoTA on GenEval benchmark. These innovations enable efficient model scaling across different compute budgets while maintaining high quality, making high-quality image generation more accessible. Our code and pre-trained models are released.

Jian Lu, Shikhar Srivastava, Junyu Chen et al.

With the advent of multi-modal large language models (MLLMs), datasets used for visual question answering (VQA) and referring expression comprehension have seen a resurgence. However, the most popular datasets used to evaluate MLLMs are some of the earliest ones created, and they have many known problems, including extreme bias, spurious correlations, and an inability to permit fine-grained analysis. In this paper, we pioneer evaluating recent MLLMs (LLaVA 1.5, LLaVA-NeXT, BLIP2, InstructBLIP, GPT-4V, and GPT-4o) on datasets designed to address weaknesses in earlier ones. We assess three VQA datasets: 1) TDIUC, which permits fine-grained analysis on 12 question types; 2) TallyQA, which has simple and complex counting questions; and 3) DVQA, which requires optical character recognition for chart understanding. We also study VQDv1, a dataset that requires identifying all image regions that satisfy a given query. Our experiments reveal the weaknesses of many MLLMs that have not previously been reported. Our code is integrated into the widely used LAVIS framework for MLLM evaluation, enabling the rapid assessment of future MLLMs. Project webpage: https://kevinlujian.github.io/MLLM_Evaluations/

Junyu Chen, Norihiro Yoshida, Hiroaki Takada

Dataset licensing is currently an issue in the development of machine learning systems. And in the development of machine learning systems, the most widely used are publicly available datasets. However, since the images in the publicly available dataset are mainly obtained from the Internet, some images are not commercially available. Furthermore, developers of machine learning systems do not often care about the license of the dataset when training machine learning models with it. In summary, the licensing of datasets for machine learning systems is in a state of incompleteness in all aspects at this stage. Our investigation of two collection datasets revealed that most of the current datasets lacked licenses, and the lack of licenses made it impossible to determine the commercial availability of the datasets. Therefore, we decided to take a more scientific and systematic approach to investigate the licensing of datasets and the licensing of machine learning systems that use the dataset to make it easier and more compliant for future developers of machine learning systems.

Jingyang Lin, Junyu Chen, Hanjia Lyu et al.

Diagnosis of adverse neonatal outcomes is crucial for preterm survival since it enables doctors to provide timely treatment. Machine learning (ML) algorithms have been demonstrated to be effective in predicting adverse neonatal outcomes. However, most previous ML-based methods have only focused on predicting a single outcome, ignoring the potential correlations between different outcomes, and potentially leading to suboptimal results and overfitting issues. In this work, we first analyze the correlations between three adverse neonatal outcomes and then formulate the diagnosis of multiple neonatal outcomes as a multi-task learning (MTL) problem. We then propose an MTL framework to jointly predict multiple adverse neonatal outcomes. In particular, the MTL framework contains shared hidden layers and multiple task-specific branches. Extensive experiments have been conducted using Electronic Health Records (EHRs) from 121 preterm neonates. Empirical results demonstrate the effectiveness of the MTL framework. Furthermore, the feature importance is analyzed for each neonatal outcome, providing insights into model interpretability.

Junyu Chen, Dongyun Zou, Wenkun He et al.

We present DC-AE 1.5, a new family of deep compression autoencoders for high-resolution diffusion models. Increasing the autoencoder's latent channel number is a highly effective approach for improving its reconstruction quality. However, it results in slow convergence for diffusion models, leading to poorer generation quality despite better reconstruction quality. This issue limits the quality upper bound of latent diffusion models and hinders the employment of autoencoders with higher spatial compression ratios. We introduce two key innovations to address this challenge: i) Structured Latent Space, a training-based approach to impose a desired channel-wise structure on the latent space with front latent channels capturing object structures and latter latent channels capturing image details; ii) Augmented Diffusion Training, an augmented diffusion training strategy with additional diffusion training objectives on object latent channels to accelerate convergence. With these techniques, DC-AE 1.5 delivers faster convergence and better diffusion scaling results than DC-AE. On ImageNet 512x512, DC-AE-1.5-f64c128 delivers better image generation quality than DC-AE-f32c32 while being 4x faster. Code: https://github.com/dc-ai-projects/DC-Gen.

Junyu Chen, Yihua Gao, Mingyuan Ge et al.

Image-text matching is crucial for bridging the semantic gap between computer vision and natural language processing. However, existing methods still face challenges in handling high-order associations and semantic ambiguities among similar instances. These ambiguities arise from subtle differences between soft positive samples (semantically similar but incorrectly labeled) and soft negative samples (locally matched but globally inconsistent), creating matching uncertainties. Furthermore, current methods fail to fully utilize the neighborhood relationships among semantically similar instances within training batches, limiting the model's ability to learn high-order shared knowledge. This paper proposes the Ambiguity-Aware and High-order Relation learning framework (AAHR) to address these issues. AAHR constructs a unified representation space through dynamic clustering prototype contrastive learning, effectively mitigating the soft positive sample problem. The framework introduces global and local feature extraction mechanisms and an adaptive aggregation network, significantly enhancing full-grained semantic understanding capabilities. Additionally, AAHR employs intra-modal and inter-modal correlation matrices to investigate neighborhood relationships among sample instances thoroughly. It incorporates GNN to enhance semantic interactions between instances. Furthermore, AAHR integrates momentum contrastive learning to expand the negative sample set. These combined strategies significantly improve the model's ability to discriminate between features. Experimental results demonstrate that AAHR outperforms existing state-of-the-art methods on Flickr30K, MSCOCO, and ECCV Caption datasets, considerably improving the accuracy and efficiency of image-text matching. The code and model checkpoints for this research are available at https://github.com/Image-Text-Matching/AAHR .

Wenkun He, Yuchao Gu, Junyu Chen et al.

Existing text-to-image diffusion models excel at generating high-quality images, but face significant efficiency challenges when scaled to high resolutions, like 4K image generation. While previous research accelerates diffusion models in various aspects, it seldom handles the inherent redundancy within the latent space. To bridge this gap, this paper introduces DC-Gen, a general framework that accelerates text-to-image diffusion models by leveraging a deeply compressed latent space. Rather than a costly training-from-scratch approach, DC-Gen uses an efficient post-training pipeline to preserve the quality of the base model. A key challenge in this paradigm is the representation gap between the base model's latent space and a deeply compressed latent space, which can lead to instability during direct fine-tuning. To overcome this, DC-Gen first bridges the representation gap with a lightweight embedding alignment training. Once the latent embeddings are aligned, only a small amount of LoRA fine-tuning is needed to unlock the base model's inherent generation quality. We verify DC-Gen's effectiveness on SANA and FLUX.1-Krea. The resulting DC-Gen-SANA and DC-Gen-FLUX models achieve quality comparable to their base models but with a significant speedup. Specifically, DC-Gen-FLUX reduces the latency of 4K image generation by 53x on the NVIDIA H100 GPU. When combined with NVFP4 SVDQuant, DC-Gen-FLUX generates a 4K image in just 3.5 seconds on a single NVIDIA 5090 GPU, achieving a total latency reduction of 138x compared to the base FLUX.1-Krea model. Code: https://github.com/dc-ai-projects/DC-Gen.

Senyao Li, Zhigang Zuo, Haozhao Wang et al.

Collaborative edge-cloud frameworks have emerged as the main- stream paradigm for mobile automation, mitigating the latency and privacy risks inherent to monolithic cloud agents. However, existing approaches centralize administration in the cloud while relegating the device to passive execution, inducing a cognitive lag regard- ing real-time UI dynamics. To tackle this, we introduce AdecPilot by applying the principle of administrative decentralization to the edge-cloud multi-agent framework, which redefines edge agency by decoupling high-level strategic designing from tactical grounding. AdecPilot integrates a UI-agnostic cloud designer generating ab- stract milestones with a bimodal edge team capable of autonomous tactical planning and self-correction without cloud intervention. Furthermore, AdecPilot employs a Hierarchical Implicit Termi- nation protocol to enforce deterministic stops and prevent post- completion hallucinations. Extensive experiments demonstrate pro- posed approach improves task success rate by 21.7% while reducing cloud token consumption by 37.5% against EcoAgent and decreas- ing end to end latency by 88.9% against CORE. The source code is available at https://anonymous.4open.science/r/Anonymous_code- B8AB.

Junyu Chen, Long Shi, Badong Chen

Graph Neural Networks (GNNs) have exhibited remarkable efficacy in learning from multi-view graph data. In the framework of multi-view graph neural networks, a critical challenge lies in effectively combining diverse views, where each view has distinct graph structure features (GSFs). Existing approaches to this challenge primarily focus on two aspects: 1) prioritizing the most important GSFs, 2) utilizing GNNs for feature aggregation. However, prioritizing the most important GSFs can lead to limited feature diversity, and existing GNN-based aggregation strategies equally treat each view without considering view quality. To address these issues, we propose a novel Multi-View Graph Neural Network with Reliable Structural Enhancement and Aggregation (RSEA-MVGNN). Firstly, we estimate view-specific uncertainty employing subjective logic. Based on this uncertainty, we design reliable structural enhancement by feature de-correlation algorithm. This approach enables each enhancement to focus on different GSFs, thereby achieving diverse feature representation in the enhanced structure. Secondly, the model learns view-specific beliefs and uncertainty as opinions, which are utilized to evaluate view quality. Based on these opinions, the model enables high-quality views to dominate GNN aggregation, thereby facilitating representation learning. Experimental results conducted on five real-world datasets demonstrate that RSEA-MVGNN outperforms several state-of-the-art GNN-based methods.

Junyu Chen, Wenkun He, Yuchao Gu et al.

We introduce DC-VideoGen, a post-training acceleration framework for efficient video generation. DC-VideoGen can be applied to any pre-trained video diffusion model, improving efficiency by adapting it to a deep compression latent space with lightweight fine-tuning. The framework builds on two key innovations: (i) a Deep Compression Video Autoencoder with a novel chunk-causal temporal design that achieves 32x/64x spatial and 4x temporal compression while preserving reconstruction quality and generalization to longer videos; and (ii) AE-Adapt-V, a robust adaptation strategy that enables rapid and stable transfer of pre-trained models into the new latent space. Adapting the pre-trained Wan-2.1-14B model with DC-VideoGen requires only 10 GPU days on the NVIDIA H100 GPU. The accelerated models achieve up to 14.8x lower inference latency than their base counterparts without compromising quality, and further enable 2160x3840 video generation on a single GPU. Code: https://github.com/dc-ai-projects/DC-VideoGen.

Enze Xie, Junsong Chen, Junyu Chen et al.

We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096$\times$4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8$\times$, we trained an AE that can compress images 32$\times$, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024$\times$1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released.

Xinhang Li, Qing Guo, Junyu Chen et al.

With increasing urban traffic complexity, Traffic Signal Control (TSC) is essential for optimizing traffic flow and improving road safety. Large Language Models (LLMs) emerge as promising approaches for TSC. However, they are prone to hallucinations in emergencies, leading to unreliable decisions that may cause substantial delays for emergency vehicles. Moreover, diverse intersection types present substantial challenges for traffic state encoding and cross-intersection training, limiting generalization across heterogeneous intersections. Therefore, this paper proposes Retrieval Augmented Generation (RAG)-enhanced distributed LLM agents with Emergency response for Generalizable TSC (REG-TSC). Firstly, this paper presents an emergency-aware reasoning framework, which dynamically adjusts reasoning depth based on the emergency scenario and is equipped with a novel Reviewer-based Emergency RAG (RERAG) to distill specific knowledge and guidance from historical cases, enhancing the reliability and rationality of agents' emergency decisions. Secondly, this paper designs a type-agnostic traffic representation and proposes a Reward-guided Reinforced Refinement (R3) for heterogeneous intersections. R3 adaptively samples training experience from diverse intersections with environment feedback-based priority and fine-tunes LLM agents with a designed reward-weighted likelihood loss, guiding REG-TSC toward high-reward policies across heterogeneous intersections. On three real-world road networks with 17 to 177 heterogeneous intersections, extensive experiments show that REG-TSC reduces travel time by 42.00%, queue length by 62.31%, and emergency vehicle waiting time by 83.16%, outperforming other state-of-the-art methods.

Junyu Chen, Yihua Gao, Mingyong Li

Image-text matching (ITM) aims to address the fundamental challenge of aligning visual and textual modalities, which inherently differ in their representations, continuous, high-dimensional image features vs. discrete, structured text. We propose a novel framework that bridges the modality gap by leveraging multimodal large language models (MLLMs) as visual semantic parsers. By generating rich Visual Semantic Descriptions (VSD), MLLMs provide semantic anchor that facilitate cross-modal alignment. Our approach combines: (1) Instance-level alignment by fusing visual features with VSD to enhance the linguistic expressiveness of image representations, and (2) Prototype-level alignment through VSD clustering to ensure category-level consistency. These modules can be seamlessly integrated into existing ITM models. Extensive experiments on Flickr30K and MSCOCO demonstrate substantial performance improvements. The approach also exhibits remarkable zero-shot generalization to cross-domain tasks, including news and remote sensing ITM. The code and model checkpoints are available at https://github.com/Image-Text-Matching/VSD.

Junyu Chen, Junzhuo Li, Zhen Peng et al.

Quantization and fine-tuning are crucial for deploying large language models (LLMs) on resource-constrained edge devices. However, fine-tuning quantized models presents significant challenges, primarily stemming from: First, the mismatch in data types between the low-precision quantized weights (e.g., 4-bit) and the high-precision adaptation weights (e.g., 16-bit). This mismatch limits the computational efficiency advantage offered by quantized weights during inference. Second, potential accuracy degradation when merging these high-precision adaptation weights into the low-precision quantized weights, as the adaptation weights often necessitate approximation or truncation. Third, as far as we know, no existing methods support the lossless merging of adaptation while adjusting all quantized weights. To address these challenges, we introduce lossless ternary adaptation for quantization-aware fine-tuning (LoTA-QAF). This is a novel fine-tuning method specifically designed for quantized LLMs, enabling the lossless merging of ternary adaptation weights into quantized weights and the adjustment of all quantized weights. LoTA-QAF operates through a combination of: i) A custom-designed ternary adaptation (TA) that aligns ternary weights with the quantization grid and uses these ternary weights to adjust quantized weights. ii) A TA-based mechanism that enables the lossless merging of adaptation weights. iii) Ternary signed gradient descent (t-SignSGD) for updating the TA weights. We apply LoTA-QAF to Llama-3.1/3.3 and Qwen-2.5 model families and validate its effectiveness on several downstream tasks. On the MMLU benchmark, our method effectively recovers performance for quantized models, surpassing 16-bit LoRA by up to 5.14\%. For task-specific fine-tuning, 16-bit LoRA achieves superior results, but LoTA-QAF still outperforms other methods. Code: github.com/KingdalfGoodman/LoTA-QAF.

Xiao Yu, Ruize Xu, Chengyuan Xue et al.

A reliable resume-job matching system helps a company find suitable candidates from a pool of resumes and helps a job seeker find relevant jobs from a list of job posts. While recent advances in embedding-based methods such as ConFit and ConFit v2 can efficiently retrieve candidates at scale, the lack of controllability and explainability limits their real-world adaptations. LLM-based re-rankers can address these limitations through reasoning, but existing training recipes are developed on short-document benchmarks and do not account for noise in real-world recruiting data. In this work, we first conduct a systematic analysis over the LLM re-ranker training pipeline for person-job fit, covering inference algorithm design, RL algorithm selection, data processing, and SFT distillation. We find that using multi-pass re-ranking, training with listwise RL objectives, removing noisy samples, and distilling from a stronger LLM before RL significantly improves re-ranking performance. We then aggregate these findings to train ConFit v3 with Qwen3-8B and Qwen3-32B on real-world person-job fit datasets, and find significant improvements over existing best person-job fit systems as well as strong LLMs such as GPT-5 and Claude Opus-4.5. We hope our findings provide useful insights for future research on adapting LLM-based re-rankers to person-job fit systems.

Zifan Wang, Junyu Chen, Ziqing Chen et al.

This paper presents GenH2R, a framework for learning generalizable vision-based human-to-robot (H2R) handover skills. The goal is to equip robots with the ability to reliably receive objects with unseen geometry handed over by humans in various complex trajectories. We acquire such generalizability by learning H2R handover at scale with a comprehensive solution including procedural simulation assets creation, automated demonstration generation, and effective imitation learning. We leverage large-scale 3D model repositories, dexterous grasp generation methods, and curve-based 3D animation to create an H2R handover simulation environment named \simabbns, surpassing the number of scenes in existing simulators by three orders of magnitude. We further introduce a distillation-friendly demonstration generation method that automatically generates a million high-quality demonstrations suitable for learning. Finally, we present a 4D imitation learning method augmented by a future forecasting objective to distill demonstrations into a visuo-motor handover policy. Experimental evaluations in both simulators and the real world demonstrate significant improvements (at least +10\% success rate) over baselines in all cases. The project page is https://GenH2R.github.io/.

Zifan Wang, Zhuorui Ye, Haoran Wu et al.

We study a new problem of semantic complete scene forecasting (SCSF) in this work. Given a 4D dynamic point cloud sequence, our goal is to forecast the complete scene corresponding to the future next frame along with its semantic labels. To tackle this challenging problem, we properly model the synergetic relationship between future forecasting and semantic scene completion through a novel network named SCSFNet. SCSFNet leverages a hybrid geometric representation for high-resolution complete scene forecasting. To leverage multi-frame observation as well as the understanding of scene dynamics to ease the completion task, SCSFNet introduces an attention-based skip connection scheme. To ease the need to model occlusion variations and to better focus on the occluded part, SCSFNet utilizes auxiliary visibility grids to guide the forecasting task. To evaluate the effectiveness of SCSFNet, we conduct experiments on various benchmarks including two large-scale indoor benchmarks we contributed and the outdoor SemanticKITTI benchmark. Extensive experiments show SCSFNet outperforms baseline methods on multiple metrics by a large margin, and also prove the synergy between future forecasting and semantic scene completion.

Junsong Chen, Yuyang Zhao, Jincheng Yu et al.

We introduce SANA-Video, a small diffusion model that can efficiently generate videos up to 720x1280 resolution and minute-length duration. SANA-Video synthesizes high-resolution, high-quality and long videos with strong text-video alignment at a remarkably fast speed, deployable on RTX 5090 GPU. Two core designs ensure our efficient, effective and long video generation: (1) Linear DiT: We leverage linear attention as the core operation, which is more efficient than vanilla attention given the large number of tokens processed in video generation. (2) Constant-Memory KV cache for Block Linear Attention: we design block-wise autoregressive approach for long video generation by employing a constant-memory state, derived from the cumulative properties of linear attention. This KV cache provides the Linear DiT with global context at a fixed memory cost, eliminating the need for a traditional KV cache and enabling efficient, minute-long video generation. In addition, we explore effective data filters and model training strategies, narrowing the training cost to 12 days on 64 H100 GPUs, which is only 1% of the cost of MovieGen. Given its low cost, SANA-Video achieves competitive performance compared to modern state-of-the-art small diffusion models (e.g., Wan 2.1-1.3B and SkyReel-V2-1.3B) while being 16x faster in measured latency. Moreover, SANA-Video can be deployed on RTX 5090 GPUs with NVFP4 precision, accelerating the inference speed of generating a 5-second 720p video from 71s to 29s (2.4x speedup). In summary, SANA-Video enables low-cost, high-quality video generation.

Yuxian Gu, Qinghao Hu, Shang Yang et al.

We present Jet-Nemotron, a new family of hybrid-architecture language models, which matches or exceeds the accuracy of leading full-attention models while significantly improving generation throughput. Jet-Nemotron is developed using Post Neural Architecture Search (PostNAS), a novel neural architecture exploration pipeline that enables efficient model design. Unlike prior approaches, PostNAS begins with a pre-trained full-attention model and freezes its MLP weights, allowing efficient exploration of attention block designs. The pipeline includes four key components: (1) learning optimal full-attention layer placement and elimination, (2) linear attention block selection, (3) designing new attention blocks, and (4) performing hardware-aware hyperparameter search. Our Jet-Nemotron-2B model achieves comparable or superior accuracy to Qwen3, Qwen2.5, Gemma3, and Llama3.2 across a comprehensive suite of benchmarks while delivering up to 53.6x generation throughput speedup and 6.1x prefilling speedup. It also achieves higher accuracy on MMLU and MMLU-Pro than recent advanced MoE full-attention models, such as DeepSeek-V3-Small and Moonlight, despite their larger scale with 15B total and 2.2B activated parameters.

Junyu Chen, Shuwen Wei, Joel Honkamaa et al.

Medical image challenges have played a transformative role in advancing the field, catalyzing algorithmic innovation and establishing new performance standards across diverse clinical applications. Image registration, a foundational task in neuroimaging pipelines, has similarly benefited from the Learn2Reg initiative. Building on this foundation, we introduce the Large-scale Unsupervised Brain MRI Image Registration (LUMIR) challenge, a next-generation benchmark designed to assess and advance unsupervised brain MRI registration. Distinct from prior challenges that leveraged anatomical label maps for supervision, LUMIR removes this dependency by providing over 4,000 preprocessed T1-weighted brain MRIs for training without any label maps, encouraging biologically plausible deformation modeling through self-supervision. In addition to evaluating performance on 590 held-out test subjects, LUMIR introduces a rigorous suite of zero-shot generalization tasks, spanning out-of-domain imaging modalities (e.g., FLAIR, T2-weighted, T2*-weighted), disease populations (e.g., Alzheimer's disease), acquisition protocols (e.g., 9.4T MRI), and species (e.g., macaque brains). A total of 1,158 subjects and over 4,000 image pairs were included for evaluation. Performance was assessed using both segmentation-based metrics (Dice coefficient, 95th percentile Hausdorff distance) and landmark-based registration accuracy (target registration error). Across both in-domain and zero-shot tasks, deep learning-based methods consistently achieved state-of-the-art accuracy while producing anatomically plausible deformation fields. The top-performing deep learning-based models demonstrated diffeomorphic properties and inverse consistency, outperforming several leading optimization-based methods, and showing strong robustness to most domain shifts, the exception being a drop in performance on out-of-domain contrasts.

Junyu Chen, Yihao Liu, Shuwen Wei et al.

Understanding the uncertainty inherent in deep learning-based image registration models has been an ongoing area of research. Existing methods have been developed to quantify both transformation and appearance uncertainties related to the registration process, elucidating areas where the model may exhibit ambiguity regarding the generated deformation. However, our study reveals that neither uncertainty effectively estimates the potential errors when the registration model is used for label propagation. Here, we propose a novel framework to concurrently estimate both the epistemic and aleatoric segmentation uncertainties for image registration. To this end, we implement a compact deep neural network (DNN) designed to transform the appearance discrepancy in the warping into aleatoric segmentation uncertainty by minimizing a negative log-likelihood loss function. Furthermore, we present epistemic segmentation uncertainty within the label propagation process as the entropy of the propagated labels. By introducing segmentation uncertainty along with existing methods for estimating registration uncertainty, we offer vital insights into the potential uncertainties at different stages of image registration. We validated our proposed framework using publicly available datasets, and the results prove that the segmentation uncertainties estimated with the proposed method correlate well with errors in label propagation, all while achieving superior registration performance.

Zhangxing Bian, Ahmed Alshareef, Shuwen Wei et al.

Magnetic Resonance Imaging with tagging (tMRI) has long been utilized for quantifying tissue motion and strain during deformation. However, a phenomenon known as tag fading, a gradual decrease in tag visibility over time, often complicates post-processing. The first contribution of this study is to model tag fading by considering the interplay between $T_1$ relaxation and the repeated application of radio frequency (RF) pulses during serial imaging sequences. This is a factor that has been overlooked in prior research on tMRI post-processing. Further, we have observed an emerging trend of utilizing raw tagged MRI within a deep learning-based (DL) registration framework for motion estimation. In this work, we evaluate and analyze the impact of commonly used image similarity objectives in training DL registrations on raw tMRI. This is then compared with the Harmonic Phase-based approach, a traditional approach which is claimed to be robust to tag fading. Our findings, derived from both simulated images and an actual phantom scan, reveal the limitations of various similarity losses in raw tMRI and emphasize caution in registration tasks where image intensity changes over time.

Yecheng Wu, Junyu Chen, Zhuoyang Zhang et al.

We introduce DC-AR, a novel masked autoregressive (AR) text-to-image generation framework that delivers superior image generation quality with exceptional computational efficiency. Due to the tokenizers' limitations, prior masked AR models have lagged behind diffusion models in terms of quality or efficiency. We overcome this limitation by introducing DC-HT - a deep compression hybrid tokenizer for AR models that achieves a 32x spatial compression ratio while maintaining high reconstruction fidelity and cross-resolution generalization ability. Building upon DC-HT, we extend MaskGIT and create a new hybrid masked autoregressive image generation framework that first produces the structural elements through discrete tokens and then applies refinements via residual tokens. DC-AR achieves state-of-the-art results with a gFID of 5.49 on MJHQ-30K and an overall score of 0.69 on GenEval, while offering 1.5-7.9x higher throughput and 2.0-3.5x lower latency compared to prior leading diffusion and autoregressive models.

Junyu Chen, Shuwen Wei, Yihao Liu et al.

Recent advances in deep learning-based medical image registration have shown that training deep neural networks~(DNNs) does not necessarily require medical images. Previous work showed that DNNs trained on randomly generated images with carefully designed noise and contrast properties can still generalize well to unseen medical data. Building on this insight, we propose using registration between random images as a proxy task for pretraining a foundation model for image registration. Empirical results show that our pretraining strategy improves registration accuracy, reduces the amount of domain-specific data needed to achieve competitive performance, and accelerates convergence during downstream training, thereby enhancing computational efficiency.

Qing Guo, Xinhang Li, Junyu Chen et al.

Traffic signal control (TSC) is a core component of intelligent transportation systems (ITS), aiming to reduce congestion, emissions, and travel time. Recent approaches based on reinforcement learning (RL) and large language models (LLMs) have improved adaptivity, but still suffer from limited interpretability, insufficient interaction data, and weak generalization to heterogeneous intersections. This paper proposes CuraLight, an LLM-centered framework where an RL agent assists the fine-tuning of an LLM-based traffic signal controller. The RL agent explores traffic environments and generates high-quality interaction trajectories, which are converted into prompt-response pairs for imitation fine-tuning. A multi-LLM ensemble deliberation system further evaluates candidate signal timing actions through structured debate, providing preference-aware supervision signals for training. Experiments conducted in SUMO across heterogeneous real-world networks from Jinan, Hangzhou, and Yizhuang demonstrate that CuraLight consistently outperforms state-of-the-art baselines, reducing average travel time by 5.34 percent, average queue length by 5.14 percent, and average waiting time by 7.02 percent. The results highlight the effectiveness of combining RL-assisted exploration with deliberation-based data curation for scalable and interpretable traffic signal control.

Wanshu Nie, Sujay V. Kumar, Junyu Chen et al.

Recent advances in machine learning such as Long Short-Term Memory (LSTM) models and Transformers have been widely adopted in hydrological applications, demonstrating impressive performance amongst deep learning models and outperforming physical models in various tasks. However, their superiority in predicting land surface states such as terrestrial water storage (TWS) that are dominated by many factors such as natural variability and human driven modifications remains unclear. Here, using the open-access, globally representative HydroGlobe dataset - comprising a baseline version derived solely from a land surface model simulation and an advanced version incorporating multi-source remote sensing data assimilation - we show that linear regression is a robust benchmark, outperforming the more complex LSTM and Temporal Fusion Transformer for TWS prediction. Our findings highlight the importance of including traditional statistical models as benchmarks when developing and evaluating deep learning models. Additionally, we emphasize the critical need to establish globally representative benchmark datasets that capture the combined impact of natural variability and human interventions.

Yihao Liu, Junyu Chen, Lianrui Zuo et al.

Objective: Deep learning-based deformable image registration has achieved strong accuracy, but remains sensitive to variations in input image characteristics such as artifacts, field-of-view mismatch, or modality difference. We aim to develop a general training paradigm that improves the robustness and generalizability of registration networks. Methods: We introduce surrogate supervision, which decouples the input domain from the supervision domain by applying estimated spatial transformations to surrogate images. This allows training on heterogeneous inputs while ensuring supervision is computed in domains where similarity is well defined. We evaluate the framework through three representative applications: artifact-robust brain MR registration, mask-agnostic lung CT registration, and multi-modal MR registration. Results: Across tasks, surrogate supervision demonstrated strong resilience to input variations including inhomogeneity field, inconsistent field-of-view, and modality differences, while maintaining high performance on well-curated data. Conclusions: Surrogate supervision provides a principled framework for training robust and generalizable deep learning-based registration models without increasing complexity. Significance: Surrogate supervision offers a practical pathway to more robust and generalizable medical image registration, enabling broader applicability in diverse biomedical imaging scenarios.

Lasse Hansen, Wiebke Heyer, Christoph Großbröhmer et al.

Medical image registration is critical for clinical applications, and fair benchmarking of different methods is essential for monitoring ongoing progress. To date, the Learn2Reg 2020-2023 challenges have released several complementary datasets and established metrics for evaluations. However, these editions did not capture all aspects of the registration problem, particularly in terms of modality diversity and task complexity. To address these limitations, the 2024 edition introduces three new tasks, including large-scale multi-modal registration and unsupervised inter-subject brain registration, as well as the first microscopy-focused benchmark within Learn2Reg. The new datasets also inspired new method developments, including invertibility constraints, pyramid features, keypoints alignment and instance optimisation.

Reuben Dorent, Laura Rigolo, Colin P. Galvin et al.

Accurate intraoperative image guidance is critical for achieving maximal safe resection in brain tumor surgery, yet neuronavigation systems based on preoperative MRI lose accuracy during the procedure due to brain shift. Aligning post-resection intraoperative ultrasound (iUS) with preoperative MRI can restore spatial accuracy by estimating brain shift deformations, but it remains a challenging problem given the large anatomical and topological changes and substantial modality intensity gap. The ReMIND2Reg 2025 Challenge provides the largest public benchmark for this task, built upon the ReMIND dataset. It offers 99 training cases, 5 validation cases, and 10 private test cases comprising paired 3D ceT1 MRI, T2 MRI, and post-resection 3D iUS volumes. Data are provided without annotations for training, while validation and test performance are evaluated on manually annotated anatomical landmarks. Metrics include target registration error (TRE), robustness to worst-case landmark misalignment (TRE30), and runtime. By establishing a standardized evaluation framework for this clinically critical and technically complex problem, ReMIND2Reg aims to accelerate the development of robust, generalizable, and clinically deployable multimodal registration algorithms for image-guided neurosurgery.

Shijia Zhang, Xiyu Ding, Brian Caffo et al.

Early diagnosis of Alzheimer's disease (AD) is critical for intervention before irreversible neurodegeneration occurs. Structural MRI (sMRI) is widely used for AD diagnosis, but conventional deep learning approaches primarily rely on intensity-based features, which require large datasets to capture subtle structural changes. Jacobian determinant maps (JSM) provide complementary information by encoding localized brain deformations, yet existing multimodal fusion strategies fail to fully integrate these features with sMRI. We propose a cross-attention fusion framework to model the intrinsic relationship between sMRI intensity and JSM-derived deformations for AD classification. Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, we compare cross-attention, pairwise self-attention, and bottleneck attention with four pre-trained 3D image encoders. Cross-attention fusion achieves superior performance, with mean ROC-AUC scores of 0.903 (+/-0.033) for AD vs. cognitively normal (CN) and 0.692 (+/-0.061) for mild cognitive impairment (MCI) vs. CN. Despite its strong performance, our model remains highly efficient, with only 1.56 million parameters--over 40 times fewer than ResNet-34 (63M) and Swin UNETR (61.98M). These findings demonstrate the potential of cross-attention fusion for improving AD diagnosis while maintaining computational efficiency.