Ge Yang

William Shen, Ge Yang, Alan Yu et al. · stanford

Self-supervised and language-supervised image models contain rich knowledge of the world that is important for generalization. Many robotic tasks, however, require a detailed understanding of 3D geometry, which is often lacking in 2D image features. This work bridges this 2D-to-3D gap for robotic manipulation by leveraging distilled feature fields to combine accurate 3D geometry with rich semantics from 2D foundation models. We present a few-shot learning method for 6-DOF grasping and placing that harnesses these strong spatial and semantic priors to achieve in-the-wild generalization to unseen objects. Using features distilled from a vision-language model, CLIP, we present a way to designate novel objects for manipulation via free-text natural language, and demonstrate its ability to generalize to unseen expressions and novel categories of objects.

Gabriel B Margolis, Ge Yang, Kartik Paigwar et al. · deepmind

Agile maneuvers such as sprinting and high-speed turning in the wild are challenging for legged robots. We present an end-to-end learned controller that achieves record agility for the MIT Mini Cheetah, sustaining speeds up to 3.9 m/s. This system runs and turns fast on natural terrains like grass, ice, and gravel and responds robustly to disturbances. Our controller is a neural network trained in simulation via reinforcement learning and transferred to the real world. The two key components are (i) an adaptive curriculum on velocity commands and (ii) an online system identification strategy for sim-to-real transfer leveraged from prior work. Videos of the robot's behaviors are available at: https://agility.csail.mit.edu/

Xuxin Cheng, Jialong Li, Shiqi Yang et al.

Teleoperation serves as a powerful method for collecting on-robot data essential for robot learning from demonstrations. The intuitiveness and ease of use of the teleoperation system are crucial for ensuring high-quality, diverse, and scalable data. To achieve this, we propose an immersive teleoperation system Open-TeleVision that allows operators to actively perceive the robot's surroundings in a stereoscopic manner. Additionally, the system mirrors the operator's arm and hand movements on the robot, creating an immersive experience as if the operator's mind is transmitted to a robot embodiment. We validate the effectiveness of our system by collecting data and training imitation learning policies on four long-horizon, precise tasks (Can Sorting, Can Insertion, Folding, and Unloading) for 2 different humanoid robots and deploy them in the real world. The system is open-sourced at: https://robot-tv.github.io/

Kaijie Zhu, Jindong Wang, Xixu Hu et al.

Deep neural networks are susceptible to adversarial examples, posing a significant security risk in critical applications. Adversarial Training (AT) is a well-established technique to enhance adversarial robustness, but it often comes at the cost of decreased generalization ability. This paper proposes Robustness Critical Fine-Tuning (RiFT), a novel approach to enhance generalization without compromising adversarial robustness. The core idea of RiFT is to exploit the redundant capacity for robustness by fine-tuning the adversarially trained model on its non-robust-critical module. To do so, we introduce module robust criticality (MRC), a measure that evaluates the significance of a given module to model robustness under worst-case weight perturbations. Using this measure, we identify the module with the lowest MRC value as the non-robust-critical module and fine-tune its weights to obtain fine-tuned weights. Subsequently, we linearly interpolate between the adversarially trained weights and fine-tuned weights to derive the optimal fine-tuned model weights. We demonstrate the efficacy of RiFT on ResNet18, ResNet34, and WideResNet34-10 models trained on CIFAR10, CIFAR100, and Tiny-ImageNet datasets. Our experiments show that \method can significantly improve both generalization and out-of-distribution robustness by around 1.5% while maintaining or even slightly enhancing adversarial robustness. Code is available at https://github.com/microsoft/robustlearn.

Timur Garipov, Sebastiaan De Peuter, Ge Yang et al.

High training costs of generative models and the need to fine-tune them for specific tasks have created a strong interest in model reuse and composition. A key challenge in composing iterative generative processes, such as GFlowNets and diffusion models, is that to realize the desired target distribution, all steps of the generative process need to be coordinated, and satisfy delicate balance conditions. In this work, we propose Compositional Sculpting: a general approach for defining compositions of iterative generative processes. We then introduce a method for sampling from these compositions built on classifier guidance. We showcase ways to accomplish compositional sculpting in both GFlowNets and diffusion models. We highlight two binary operations $\unicode{x2014}$ the harmonic mean ($p_1 \otimes p_2$) and the contrast ($p_1 \unicode{x25D1}\,p_2$) between pairs, and the generalization of these operations to multiple component distributions. We offer empirical results on image and molecular generation tasks.

Meituan LongCat Team, Bin Xiao, Chao Wang et al.

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

Ge Yang, Anurag Ajay, Pulkit Agrawal

Value approximation using deep neural networks is at the heart of off-policy deep reinforcement learning, and is often the primary module that provides learning signals to the rest of the algorithm. While multi-layer perceptron networks are universal function approximators, recent works in neural kernel regression suggest the presence of a spectral bias, where fitting high-frequency components of the value function requires exponentially more gradient update steps than the low-frequency ones. In this work, we re-examine off-policy reinforcement learning through the lens of kernel regression and propose to overcome such bias via a composite neural tangent kernel. With just a single line-change, our approach, the Fourier feature networks (FFN) produce state-of-the-art performance on challenging continuous control domains with only a fraction of the compute. Faster convergence and better off-policy stability also make it possible to remove the target network without suffering catastrophic divergences, which further reduces TD}(0)'s estimation bias on a few tasks.

Jiaxing Huang, Yanfeng Zhou, Yaoru Luo et al.

Accurate segmentation of long and thin tubular structures is required in a wide variety of areas such as biology, medicine, and remote sensing. The complex topology and geometry of such structures often pose significant technical challenges. A fundamental property of such structures is their topological self-similarity, which can be quantified by fractal features such as fractal dimension (FD). In this study, we incorporate fractal features into a deep learning model by extending FD to the pixel-level using a sliding window technique. The resulting fractal feature maps (FFMs) are then incorporated as additional input to the model and additional weight in the loss function to enhance segmentation performance by utilizing the topological self-similarity. Moreover, we extend the U-Net architecture by incorporating an edge decoder and a skeleton decoder to improve boundary accuracy and skeletal continuity of segmentation, respectively. Extensive experiments on five tubular structure datasets validate the effectiveness and robustness of our approach. Furthermore, the integration of FFMs with other popular segmentation models such as HR-Net also yields performance enhancement, suggesting FFM can be incorporated as a plug-in module with different model architectures. Code and data are openly accessible at https://github.com/cbmi-group/FFM-Multi-Decoder-Network.

Yanfeng Zhou, Lingrui Li, Zichen Wang et al.

XNet introduces a wavelet-based X-shaped unified architecture for fully- and semi-supervised biomedical segmentation. So far, however, XNet still faces the limitations, including performance degradation when images lack high-frequency (HF) information, underutilization of raw images and insufficient fusion. To address these issues, we propose XNet v2, a low- and high-frequency complementary model. XNet v2 performs wavelet-based image-level complementary fusion, using fusion results along with raw images inputs three different sub-networks to construct consistency loss. Furthermore, we introduce a feature-level fusion module to enhance the transfer of low-frequency (LF) information and HF information. XNet v2 achieves state-of-the-art in semi-supervised segmentation while maintaining competitve results in fully-supervised learning. More importantly, XNet v2 excels in scenarios where XNet fails. Compared to XNet, XNet v2 exhibits fewer limitations, better results and greater universality. Extensive experiments on three 2D and two 3D datasets demonstrate the effectiveness of XNet v2. Code is available at https://github.com/Yanfeng-Zhou/XNetv2 .

Takuma Yoneda, Luzhe Sun, Ge Yang et al.

Shared autonomy is an operational concept in which a user and an autonomous agent collaboratively control a robotic system. It provides a number of advantages over the extremes of full-teleoperation and full-autonomy in many settings. Traditional approaches to shared autonomy rely on knowledge of the environment dynamics, a discrete space of user goals that is known a priori, or knowledge of the user's policy -- assumptions that are unrealistic in many domains. Recent works relax some of these assumptions by formulating shared autonomy with model-free deep reinforcement learning (RL). In particular, they no longer need knowledge of the goal space (e.g., that the goals are discrete or constrained) or environment dynamics. However, they need knowledge of a task-specific reward function to train the policy. Unfortunately, such reward specification can be a difficult and brittle process. On top of that, the formulations inherently rely on human-in-the-loop training, and that necessitates them to prepare a policy that mimics users' behavior. In this paper, we present a new approach to shared autonomy that employs a modulation of the forward and reverse diffusion process of diffusion models. Our approach does not assume known environment dynamics or the space of user goals, and in contrast to previous work, it does not require any reward feedback, nor does it require access to the user's policy during training. Instead, our framework learns a distribution over a space of desired behaviors. It then employs a diffusion model to translate the user's actions to a sample from this distribution. Crucially, we show that it is possible to carry out this process in a manner that preserves the user's control authority. We evaluate our framework on a series of challenging continuous control tasks, and analyze its ability to effectively correct user actions while maintaining their autonomy.

Yunpeng Xiao, Youpeng Zhao, Ge Yang

Morphology of mitochondria plays critical roles in mediating their physiological functions. Accurate segmentation of mitochondria from 3D electron microscopy (EM) images is essential to quantitative characterization of their morphology at the nanometer scale. Fully supervised deep learning models developed for this task achieve excellent performance but require substantial amounts of annotated data for training. However, manual annotation of EM images is laborious and time-consuming because of their large volumes, limited contrast, and low signal-to-noise ratios (SNRs). To overcome this challenge, we propose a semi-supervised deep learning model that segments mitochondria by leveraging the spatial continuity of their structural, morphological, and contextual information in both labeled and unlabeled images. We use random piecewise affine transformation to synthesize comprehensive and realistic mitochondrial morphology for augmentation of training data. Experiments on the EPFL dataset show that our model achieves performance similar as that of state-of-the-art fully supervised models but requires only ~20% of their annotated training data. Our semi-supervised model is versatile and can also accurately segment other spatially continuous structures from EM images. Data and code of this study are openly accessible at https://github.com/cbmi-group/MPP.

Zhang-Wei Hong, Ge Yang, Pulkit Agrawal

The dominant framework for off-policy multi-goal reinforcement learning involves estimating goal conditioned Q-value function. When learning to achieve multiple goals, data efficiency is intimately connected with the generalization of the Q-function to new goals. The de-facto paradigm is to approximate Q(s, a, g) using monolithic neural networks. To improve the generalization of the Q-function, we propose a bilinear decomposition that represents the Q-value via a low-rank approximation in the form of a dot product between two vector fields. The first vector field, f(s, a), captures the environment's local dynamics at the state s; whereas the second component, φ(s, g), captures the global relationship between the current state and the goal. We show that our bilinear decomposition scheme substantially improves data efficiency, and has superior transfer to out-of-distribution goals compared to prior methods. Empirical evidence is provided on the simulated Fetch robot task-suite and dexterous manipulation with a Shadow hand.

Ruihan Yang, Ge Yang, Xiaolong Wang

Legged robots have the potential to expand the reach of autonomy beyond paved roads. In this work, we consider the difficult problem of locomotion on challenging terrains using a single forward-facing depth camera. Due to the partial observability of the problem, the robot has to rely on past observations to infer the terrain currently beneath it. To solve this problem, we follow the paradigm in computer vision that explicitly models the 3D geometry of the scene and propose Neural Volumetric Memory (NVM), a geometric memory architecture that explicitly accounts for the SE(3) equivariance of the 3D world. NVM aggregates feature volumes from multiple camera views by first bringing them back to the ego-centric frame of the robot. We test the learned visual-locomotion policy on a physical robot and show that our approach, which explicitly introduces geometric priors during training, offers superior performance than more naïve methods. We also include ablation studies and show that the representations stored in the neural volumetric memory capture sufficient geometric information to reconstruct the scene. Our project page with videos is https://rchalyang.github.io/NVM .

Jian Zhang, Runwei Ding, Miaoju Ban et al.

Visual anomaly detection is essential and commonly used for many tasks in the field of computer vision. Recent anomaly detection datasets mainly focus on industrial automated inspection, medical image analysis and video surveillance. In order to broaden the application and research of anomaly detection in unmanned supermarkets and smart manufacturing, we introduce the supermarket goods anomaly detection (GoodsAD) dataset. It contains 6124 high-resolution images of 484 different appearance goods divided into 6 categories. Each category contains several common different types of anomalies such as deformation, surface damage and opened. Anomalies contain both texture changes and structural changes. It follows the unsupervised setting and only normal (defect-free) images are used for training. Pixel-precise ground truth regions are provided for all anomalies. Moreover, we also conduct a thorough evaluation of current state-of-the-art unsupervised anomaly detection methods. This initial benchmark indicates that some methods which perform well on the industrial anomaly detection dataset (e.g., MVTec AD), show poor performance on our dataset. This is a comprehensive, multi-object dataset for supermarket goods anomaly detection that focuses on real-world applications.

Siddharth Mishra-Sharma, Ge Yang

From the nature of dark matter to the rate of expansion of our Universe, observations of distant galaxies distorted through strong gravitational lensing have the potential to answer some of the major open questions in astrophysics. Modeling galaxy-galaxy strong lensing observations presents a number of challenges as the exact configuration of both the background source and foreground lens galaxy is unknown. A timely call, prompted by a number of upcoming surveys anticipating high-resolution lensing images, demands methods that can efficiently model lenses at their full complexity. In this work, we introduce a method that uses continuous neural fields to non-parametrically reconstruct the complex morphology of a source galaxy while simultaneously inferring a distribution over foreground lens galaxy configurations. We demonstrate the efficacy of our method through experiments on simulated data targeting high-resolution lensing images similar to those anticipated in near-future astrophysical surveys.

Lingrui Li, Yanfeng Zhou, Ge Yang

Unsupervised Domain Adaptation (UDA) is a learning technique that transfers knowledge learned in the source domain from labelled training data to the target domain with only unlabelled data. It is of significant importance to medical image segmentation because of the usual lack of labelled training data. Although extensive efforts have been made to optimize UDA techniques to improve the accuracy of segmentation models in the target domain, few studies have addressed the robustness of these models under UDA. In this study, we propose a two-stage training strategy for robust domain adaptation. In the source training stage, we utilize adversarial sample augmentation to enhance the robustness and generalization capability of the source model. And in the target training stage, we propose a novel robust pseudo-label and pseudo-boundary (PLPB) method, which effectively utilizes unlabeled target data to generate pseudo labels and pseudo boundaries that enable model self-adaptation without requiring source data. Extensive experimental results on cross-domain fundus image segmentation confirm the effectiveness and versatility of our method. Source code of this study is openly accessible at https://github.com/LinGrayy/PLPB.

Guoqing Ma, Haoyang Huang, Kun Yan et al.

We present Step-Video-T2V, a state-of-the-art text-to-video pre-trained model with 30B parameters and the ability to generate videos up to 204 frames in length. A deep compression Variational Autoencoder, Video-VAE, is designed for video generation tasks, achieving 16x16 spatial and 8x temporal compression ratios, while maintaining exceptional video reconstruction quality. User prompts are encoded using two bilingual text encoders to handle both English and Chinese. A DiT with 3D full attention is trained using Flow Matching and is employed to denoise input noise into latent frames. A video-based DPO approach, Video-DPO, is applied to reduce artifacts and improve the visual quality of the generated videos. We also detail our training strategies and share key observations and insights. Step-Video-T2V's performance is evaluated on a novel video generation benchmark, Step-Video-T2V-Eval, demonstrating its state-of-the-art text-to-video quality when compared with both open-source and commercial engines. Additionally, we discuss the limitations of current diffusion-based model paradigm and outline future directions for video foundation models. We make both Step-Video-T2V and Step-Video-T2V-Eval available at https://github.com/stepfun-ai/Step-Video-T2V. The online version can be accessed from https://yuewen.cn/videos as well. Our goal is to accelerate the innovation of video foundation models and empower video content creators.

Mengxuan Chen, Yunpu Xu, Qiuyan Sun et al.

Seasonal forecasting of summer rainfall in East Asia remains a grand challenge, as predictability at 3 to 6 month lead times is constrained by the spring predictability barrier, weak large-scale signals, and localized nonlinear convective extremes. We address this challenge with CAPES, which integrates a kilometer-resolution coupled regional model with atmosphere, land, and ocean components and a data-driven AI seasonal forecasting system. At 15 km resolution, the fused workflow combines 174 numerical members from varying start times, physics schemes, and parameter perturbations with 1,600 AI members generated from initial and physical perturbations. Using the full LineShine system, CAPES completes ten annual 1,774-member hindcasts for 2016 to 2025 within 14.6 hours, improving the mean prediction score from ECMWF's 71.8 to 75.9 and delivering a major gain in operational forecasting capability. The 1-km configuration further enables fine-scale typhoon simulation and establishes the feasibility of kilometer-scale fused ensemble forecasting on a one-week timescale.

Yiming Huang, Guole Liu, Yaoru Luo et al.

The scarcity of annotated data, particularly for rare diseases, limits the variability of training data and the range of detectable lesions, presenting a significant challenge for supervised anomaly detection in medical imaging. To solve this problem, we propose a novel unsupervised method for medical image anomaly detection: Attention-Augmented Differentiable top-k Feature Adaptation (ADFA). The method utilizes Wide-ResNet50-2 (WR50) network pre-trained on ImageNet to extract initial feature representations. To reduce the channel dimensionality while preserving relevant channel information, we employ an attention-augmented patch descriptor on the extracted features. We then apply differentiable top-k feature adaptation to train the patch descriptor, mapping the extracted feature representations to a new vector space, enabling effective detection of anomalies. Experiments show that ADFA outperforms state-of-the-art (SOTA) methods on multiple challenging medical image datasets, confirming its effectiveness in medical anomaly detection.

Ge Yang, Changyi He, Jinyang Guo et al.

Although large language models (LLMs) have demonstrated their strong intelligence ability, the high demand for computation and storage hinders their practical application. To this end, many model compression techniques are proposed to increase the efficiency of LLMs. However, current researches only validate their methods on limited models, datasets, metrics, etc, and still lack a comprehensive evaluation under more general scenarios. So it is still a question of which model compression approach we should use under a specific case. To mitigate this gap, we present the Large Language Model Compression Benchmark (LLMCBench), a rigorously designed benchmark with an in-depth analysis for LLM compression algorithms. We first analyze the actual model production requirements and carefully design evaluation tracks and metrics. Then, we conduct extensive experiments and comparison using multiple mainstream LLM compression approaches. Finally, we perform an in-depth analysis based on the evaluation and provide useful insight for LLM compression design. We hope our LLMCBench can contribute insightful suggestions for LLM compression algorithm design and serve as a foundation for future research. Our code is available at https://github.com/AboveParadise/LLMCBench.

Yaoru Luo, Guole Liu, Yuanhao Guo et al.

Supervised training of deep neural networks (DNNs) by noisy labels has been studied extensively in image classification but much less in image segmentation. Our understanding of the learning behavior of DNNs trained by noisy segmentation labels remains limited. We address this deficiency in both binary segmentation of biological microscopy images and multi-class segmentation of natural images. We classify segmentation labels according to their noise transition matrices (NTMs) and compare performance of DNNs trained by different types of labels. When we randomly sample a small fraction (e.g., 10%) or flip a large fraction (e.g., 90%) of the ground-truth labels to train DNNs, their segmentation performance remains largely unchanged. This indicates that DNNs learn structures hidden in labels rather than pixel-level labels per se in their supervised training for semantic segmentation. We call these hidden structures meta-structures. When labels with different perturbations to the meta-structures are used to train DNNs, their performance in feature extraction and segmentation degrades consistently. In contrast, addition of meta-structure information substantially improves performance of an unsupervised model in binary semantic segmentation. We formulate meta-structures mathematically as spatial density distributions. We show theoretically and experimentally how this formulation explains key observed learning behavior of DNNs.

Xuxin Cheng, Yandong Ji, Junming Chen et al.

Can we enable humanoid robots to generate rich, diverse, and expressive motions in the real world? We propose to learn a whole-body control policy on a human-sized robot to mimic human motions as realistic as possible. To train such a policy, we leverage the large-scale human motion capture data from the graphics community in a Reinforcement Learning framework. However, directly performing imitation learning with the motion capture dataset would not work on the real humanoid robot, given the large gap in degrees of freedom and physical capabilities. Our method Expressive Whole-Body Control (Exbody) tackles this problem by encouraging the upper humanoid body to imitate a reference motion, while relaxing the imitation constraint on its two legs and only requiring them to follow a given velocity robustly. With training in simulation and Sim2Real transfer, our policy can control a humanoid robot to walk in different styles, shake hands with humans, and even dance with a human in the real world. We conduct extensive studies and comparisons on diverse motions in both simulation and the real world to show the effectiveness of our approach.

Mazeyu Ji, Xuanbin Peng, Fangchen Liu et al.

This paper tackles the challenge of enabling real-world humanoid robots to perform expressive and dynamic whole-body motions while maintaining overall stability and robustness. We propose Advanced Expressive Whole-Body Control (Exbody2), a method for producing whole-body tracking controllers that are trained on both human motion capture and simulated data and then transferred to the real world. We introduce a technique for decoupling the velocity tracking of the entire body from tracking body landmarks. We use a teacher policy to produce intermediate data that better conforms to the robot's kinematics and to automatically filter away infeasible whole-body motions. This two-step approach enabled us to produce a student policy that can be deployed on the robot that can walk, crouch, and dance. We also provide insight into the trade-off between versatility and the tracking performance on specific motions. We observed significant improvement of tracking performance after fine-tuning on a small amount of data, at the expense of the others.

Chenhao Lu, Xuxin Cheng, Jialong Li et al.

Humanoid robots require both robust lower-body locomotion and precise upper-body manipulation. While recent Reinforcement Learning (RL) approaches provide whole-body loco-manipulation policies, they lack precise manipulation with high DoF arms. In this paper, we propose decoupling upper-body control from locomotion, using inverse kinematics (IK) and motion retargeting for precise manipulation, while RL focuses on robust lower-body locomotion. We introduce PMP (Predictive Motion Priors), trained with Conditional Variational Autoencoder (CVAE) to effectively represent upper-body motions. The locomotion policy is trained conditioned on this upper-body motion representation, ensuring that the system remains robust with both manipulation and locomotion. We show that CVAE features are crucial for stability and robustness, and significantly outperforms RL-based whole-body control in precise manipulation. With precise upper-body motion and robust lower-body locomotion control, operators can remotely control the humanoid to walk around and explore different environments, while performing diverse manipulation tasks.

Ri-Zhao Qiu, Ge Yang, Weijia Zeng et al.

Scene representations using 3D Gaussian primitives have produced excellent results in modeling the appearance of static and dynamic 3D scenes. Many graphics applications, however, demand the ability to manipulate both the appearance and the physical properties of objects. We introduce Feature Splatting, an approach that unifies physics-based dynamic scene synthesis with rich semantics from vision language foundation models that are grounded by natural language. Our first contribution is a way to distill high-quality, object-centric vision-language features into 3D Gaussians, that enables semi-automatic scene decomposition using text queries. Our second contribution is a way to synthesize physics-based dynamics from an otherwise static scene using a particle-based simulator, in which material properties are assigned automatically via text queries. We ablate key techniques used in this pipeline, to illustrate the challenge and opportunities in using feature-carrying 3D Gaussians as a unified format for appearance, geometry, material properties and semantics grounded on natural language. Project website: https://feature-splatting.github.io/

Ri-Zhao Qiu, Yafei Hu, Yuchen Song et al.

An open problem in mobile manipulation is how to represent objects and scenes in a unified manner so that robots can use both for navigation and manipulation. The latter requires capturing intricate geometry while understanding fine-grained semantics, whereas the former involves capturing the complexity inherent at an expansive physical scale. In this work, we present GeFF (Generalizable Feature Fields), a scene-level generalizable neural feature field that acts as a unified representation for both navigation and manipulation that performs in real-time. To do so, we treat generative novel view synthesis as a pre-training task, and then align the resulting rich scene priors with natural language via CLIP feature distillation. We demonstrate the effectiveness of this approach by deploying GeFF on a quadrupedal robot equipped with a manipulator. We quantitatively evaluate GeFF's ability for open-vocabulary object-/part-level manipulation and show that GeFF outperforms point-based baselines in runtime and storage-accuracy trade-offs, with qualitative examples of semantics-aware navigation and articulated object manipulation.

Ri-Zhao Qiu, Shiqi Yang, Xuxin Cheng et al.

Training manipulation policies for humanoid robots with diverse data enhances their robustness and generalization across tasks and platforms. However, learning solely from robot demonstrations is labor-intensive, requiring expensive tele-operated data collection which is difficult to scale. This paper investigates a more scalable data source, egocentric human demonstrations, to serve as cross-embodiment training data for robot learning. We mitigate the embodiment gap between humanoids and humans from both the data and modeling perspectives. We collect an egocentric task-oriented dataset (PH2D) that is directly aligned with humanoid manipulation demonstrations. We then train a human-humanoid behavior policy, which we term Human Action Transformer (HAT). The state-action space of HAT is unified for both humans and humanoid robots and can be differentiably retargeted to robot actions. Co-trained with smaller-scale robot data, HAT directly models humanoid robots and humans as different embodiments without additional supervision. We show that human data improves both generalization and robustness of HAT with significantly better data collection efficiency. Code and data: https://human-as-robot.github.io/

Yajvan Ravan, Adam Rashid, Alan Yu et al.

We introduce Lucid-XR, a generative data engine for creating diverse and realistic-looking multi-modal data to train real-world robotic systems. At the core of Lucid-XR is vuer, a web-based physics simulation environment that runs directly on the XR headset, enabling internet-scale access to immersive, latency-free virtual interactions without requiring specialized equipment. The complete system integrates on-device physics simulation with human-to-robot pose retargeting. Data collected is further amplified by a physics-guided video generation pipeline steerable via natural language specifications. We demonstrate zero-shot transfer of robot visual policies to unseen, cluttered, and badly lit evaluation environments, after training entirely on Lucid-XR's synthetic data. We include examples across dexterous manipulation tasks that involve soft materials, loosely bound particles, and rigid body contact. Project website: https://lucidxr.github.io

Li Xiao, Liqing Liu, Hongjun Wu et al.

Recent advances in 2D large-scale and 3D volume electron microscopy have stimulated the rapid development of nanoscale functional analysis at the tissue and organ levels. Digitizing the cell by mapping the intricate organellar networks into its physiological and pathological textures will revolutionarize the contents of cell atlases. To meet the requirements of characterizing intracellular organelles and their interactions within defined cellular cohorts at tissue level, we have developed DeepOrganelle. It adopts a lightweighted Mask2Former frameworks as a universal segmentor and is capable of segmenting and extracting organelles within different cell types, performing statistical quantitative analysis, as well as visualizing and quantifying the spatial distribution of organelle morphologies and interactions across different cell types at tissue scales. Using DeepOrganelle, we systemically perform cross-scale quantification of membrane contact sites(MCSs) dynamics across the progression of the seminiferous epithelial cycle, covering 12 distinct developmental stages and 24 statuses of germ cells. DeepOrganelle uncovers the spatiotemporal gradient of the germ cell differentiation atlas according to different types of organelles and their interactions. Noticeably, it discovers a waved pattern of mitochondria(Mito)-endoplasmic reticulum(ER) contact with a significant increase specifically at Stage X pachytene preceding the transition to diplotene, which aligns well with a newly reported experiment that mitochondrial metabolic proteins like PDHA2 are essential for this transition by maintaining ATP supply for double-strand break(DSB) repair. DeepOrganelle also observes a dynamic restructuring of the blood-testis barrier and stage-specific reorganization of organelle topography in Sertoli cells from preleptotene to leptotene phases of prophase I.

Ri-Zhao Qiu, Yuchen Song, Xuanbin Peng et al.

'In-the-wild' mobile manipulation aims to deploy robots in diverse real-world environments, which requires the robot to (1) have skills that generalize across object configurations; (2) be capable of long-horizon task execution in diverse environments; and (3) perform complex manipulation beyond pick-and-place. Quadruped robots with manipulators hold promise for extending the workspace and enabling robust locomotion, but existing results do not investigate such a capability. This paper proposes WildLMa with three components to address these issues: (1) adaptation of learned low-level controller for VR-enabled whole-body teleoperation and traversability; (2) WildLMa-Skill -- a library of generalizable visuomotor skills acquired via imitation learning or heuristics and (3) WildLMa-Planner -- an interface of learned skills that allow LLM planners to coordinate skills for long-horizon tasks. We demonstrate the importance of high-quality training data by achieving higher grasping success rate over existing RL baselines using only tens of demonstrations. WildLMa exploits CLIP for language-conditioned imitation learning that empirically generalizes to objects unseen in training demonstrations. Besides extensive quantitative evaluation, we qualitatively demonstrate practical robot applications, such as cleaning up trash in university hallways or outdoor terrains, operating articulated objects, and rearranging items on a bookshelf.

Jia He, Bonan Li, Ge Yang et al.

Solving 3D medical inverse problems such as image restoration and reconstruction is crucial in modern medical field. However, the curse of dimensionality in 3D medical data leads mainstream volume-wise methods to suffer from high resource consumption and challenges models to successfully capture the natural distribution, resulting in inevitable volume inconsistency and artifacts. Some recent works attempt to simplify generation in the latent space but lack the capability to efficiently model intricate image details. To address these limitations, we present Blaze3DM, a novel approach that enables fast and high-fidelity generation by integrating compact triplane neural field and powerful diffusion model. In technique, Blaze3DM begins by optimizing data-dependent triplane embeddings and a shared decoder simultaneously, reconstructing each triplane back to the corresponding 3D volume. To further enhance 3D consistency, we introduce a lightweight 3D aware module to model the correlation of three vertical planes. Then, diffusion model is trained on latent triplane embeddings and achieves both unconditional and conditional triplane generation, which is finally decoded to arbitrary size volume. Extensive experiments on zero-shot 3D medical inverse problem solving, including sparse-view CT, limited-angle CT, compressed-sensing MRI, and MRI isotropic super-resolution, demonstrate that Blaze3DM not only achieves state-of-the-art performance but also markedly improves computational efficiency over existing methods (22~40x faster than previous work).

Ao Li, Wei Fang, Hongbo Zhao et al.

In applications of diffusion models, controllable generation is of practical significance, but is also challenging. Current methods for controllable generation primarily focus on modifying the score function of diffusion models, while Mean Reverting (MR) Diffusion directly modifies the structure of the stochastic differential equation (SDE), making the incorporation of image conditions simpler and more natural. However, current training-free fast samplers are not directly applicable to MR Diffusion. And thus MR Diffusion requires hundreds of NFEs (number of function evaluations) to obtain high-quality samples. In this paper, we propose a new algorithm named MaRS (MR Sampler) to reduce the sampling NFEs of MR Diffusion. We solve the reverse-time SDE and the probability flow ordinary differential equation (PF-ODE) associated with MR Diffusion, and derive semi-analytical solutions. The solutions consist of an analytical function and an integral parameterized by a neural network. Based on this solution, we can generate high-quality samples in fewer steps. Our approach does not require training and supports all mainstream parameterizations, including noise prediction, data prediction and velocity prediction. Extensive experiments demonstrate that MR Sampler maintains high sampling quality with a speedup of 10 to 20 times across ten different image restoration tasks. Our algorithm accelerates the sampling procedure of MR Diffusion, making it more practical in controllable generation.

Ziming Liu, Leichen Wang, Ge Yang et al.

Reliable autonomous driving systems require high-definition (HD) map that contains detailed map information for planning and navigation. However, pre-build HD map requires a large cost. Visual-based Online Map Generation (OMG) has become an alternative low-cost solution to build a local HD map. Query-based BEV Transformer has been a base model for this task. This model learns HD map predictions from an initial map queries distribution which is obtained by offline optimization on training set. Besides the quality of BEV feature, the performance of this model also highly relies on the capacity of initial map query distribution. However, this distribution is limited because the limited query number. To make map predictions optimal on each test sample, it is essential to generate a suitable initial distribution for each specific scenario. This paper proposes to decompose the whole HD map distribution into a set of point representations, namely map query bank (MQBank). To build specific map query initial distributions of different scenarios, low-cost standard definition map (SD map) data is introduced as a kind of prior knowledge. Moreover, each layer of map decoder network learns instance-level map query features, which will lose detailed information of each point. However, BEV feature map is a point-level dense feature. It is important to keep point-level information in map queries when interacting with BEV feature map. This can also be solved with map query bank method. Final experiments show a new insight on SD map prior and a new record on OpenLaneV2 benchmark with 40.5%, 45.7% mAP on vehicle lane and pedestrian area.

Jiaxing Huang, Heng Guo, Le Lu et al.

Osteoporosis, characterized by reduced bone mineral density (BMD) and compromised bone microstructure, increases fracture risk in aging populations. While dual-energy X-ray absorptiometry (DXA) is the clinical standard for BMD assessment, its limited accessibility hinders diagnosis in resource-limited regions. Opportunistic computed tomography (CT) analysis has emerged as a promising alternative for osteoporosis diagnosis using existing imaging data. Current approaches, however, face three limitations: (1) underutilization of unlabeled vertebral data, (2) systematic bias from device-specific DXA discrepancies, and (3) insufficient integration of clinical knowledge such as spatial BMD distribution patterns. To address these, we propose a unified deep learning framework with three innovations. First, a self-supervised learning method using radiomic representations to leverage unlabeled CT data and preserve bone texture. Second, a Mixture of Experts (MoE) architecture with learned gating mechanisms to enhance cross-device adaptability. Third, a multi-task learning framework integrating osteoporosis diagnosis, BMD regression, and vertebra location prediction. Validated across three clinical sites and an external hospital, our approach demonstrates superior generalizability and accuracy over existing methods for opportunistic osteoporosis screening and diagnosis.

Jiaheng Zhou, Yanfeng Zhou, Wei Fang et al.

Ultrasound videos are an important form of clinical imaging data, and deep learning-based automated analysis can improve diagnostic accuracy and clinical efficiency. However, the scarcity of labeled data and the inherent challenges of video analysis have impeded the advancement of related methods. In this work, we introduce E-ViM$^3$, a data-efficient Vision Mamba network that preserves the 3D structure of video data, enhancing long-range dependencies and inductive biases to better model space-time correlations. With our design of Enclosure Global Tokens (EGT), the model captures and aggregates global features more effectively than competing methods. To further improve data efficiency, we employ masked video modeling for self-supervised pre-training, with the proposed Spatial-Temporal Chained (STC) masking strategy designed to adapt to various video scenarios. Experiments demonstrate that E-ViM$^3$ performs as the state-of-the-art in two high-level semantic analysis tasks across four datasets of varying sizes: EchoNet-Dynamic, CAMUS, MICCAI-BUV, and WHBUS. Furthermore, our model achieves competitive performance with limited labels, highlighting its potential impact on real-world clinical applications.

Hui Xie, Ge Yang, Wenjuan Gao

With the rapid development of deep learning, Deep Spiking Neural Networks (DSNNs) have emerged as promising due to their unique spike event processing and asynchronous computation. When deployed on neuromorphic chips, DSNNs offer significant power advantages over Deep Artificial Neural Networks (DANNs) and eliminate time and energy consuming multiplications due to the binary nature of spikes (0 or 1). Additionally, DSNNs excel in processing temporal information, making them potentially superior for handling temporal data compared to DANNs. However, their deep network structure and numerous parameters result in high computational costs and energy consumption, limiting real-life deployment. To enhance DSNNs efficiency, researchers have adapted methods from DANNs, such as pruning, quantization, and knowledge distillation, and developed specific techniques like reducing spike firing and pruning time steps. While previous surveys have covered DSNNs algorithms, hardware deployment, and general overviews, focused research on DSNNs compression and efficiency has been lacking. This survey addresses this gap by concentrating on efficient DSNNs and their compression methods. It begins with an exploration of DSNNs' biological background and computational units, highlighting differences from DANNs. It then delves into various compression methods, including pruning, quantization, knowledge distillation, and reducing spike firing, and concludes with suggestions for future research directions.

Takuma Yoneda, Ge Yang, Matthew R. Walter et al.

A robot's deployment environment often involves perceptual changes that differ from what it has experienced during training. Standard practices such as data augmentation attempt to bridge this gap by augmenting source images in an effort to extend the support of the training distribution to better cover what the agent might experience at test time. In many cases, however, it is impossible to know test-time distribution-shift a priori, making these schemes infeasible. In this paper, we introduce a general approach, called Invariance Through Latent Alignment (ILA), that improves the test-time performance of a visuomotor control policy in deployment environments with unknown perceptual variations. ILA performs unsupervised adaptation at deployment-time by matching the distribution of latent features on the target domain to the agent's prior experience, without relying on paired data. Although simple, we show that this idea leads to surprising improvements on a variety of challenging adaptation scenarios, including changes in lighting conditions, the content in the scene, and camera poses. We present results on calibrated control benchmarks in simulation -- the distractor control suite -- and a physical robot under a sim-to-real setup.

Xiang Fu, Ge Yang, Pulkit Agrawal et al.

Current model-based reinforcement learning methods struggle when operating from complex visual scenes due to their inability to prioritize task-relevant features. To mitigate this problem, we propose learning Task Informed Abstractions (TIA) that explicitly separates reward-correlated visual features from distractors. For learning TIA, we introduce the formalism of Task Informed MDP (TiMDP) that is realized by training two models that learn visual features via cooperative reconstruction, but one model is adversarially dissociated from the reward signal. Empirical evaluation shows that TIA leads to significant performance gains over state-of-the-art methods on many visual control tasks where natural and unconstrained visual distractions pose a formidable challenge.

Tianjie Yang, Yaoru Luo, Wei Ji et al.

Super-resolution microscopy overcomes the diffraction limit of conventional light microscopy in spatial resolution. By providing novel spatial or spatio-temporal information on biological processes at nanometer resolution with molecular specificity, it plays an increasingly important role in life sciences. However, its technical limitations require trade-offs to balance its spatial resolution, temporal resolution, and light exposure of samples. Recently, deep learning has achieved breakthrough performance in many image processing and computer vision tasks. It has also shown great promise in pushing the performance envelope of super-resolution microscopy. In this brief Review, we survey recent advances in using deep learning to enhance performance of super-resolution microscopy. We focus primarily on how deep learning ad-vances reconstruction of super-resolution images. Related key technical challenges are discussed. Despite the challenges, deep learning is set to play an indispensable and transformative role in the development of super-resolution microscopy. We conclude with an outlook on how deep learning could shape the future of this new generation of light microscopy technology.

Yaoru Luo, Guole Liu, Yuanhao Guo et al.

How deep neural networks (DNNs) learn from noisy labels has been studied extensively in image classification but much less in image segmentation. So far, our understanding of the learning behavior of DNNs trained by noisy segmentation labels remains limited. In this study, we address this deficiency in both binary segmentation of biological microscopy images and multi-class segmentation of natural images. We generate extremely noisy labels by randomly sampling a small fraction (e.g., 10%) or flipping a large fraction (e.g., 90%) of the ground truth labels. When trained with these noisy labels, DNNs provide largely the same segmentation performance as trained by the original ground truth. This indicates that DNNs learn structures hidden in labels rather than pixel-level labels per se in their supervised training for semantic segmentation. We refer to these hidden structures in labels as meta-structures. When DNNs are trained by labels with different perturbations to the meta-structure, we find consistent degradation in their segmentation performance. In contrast, incorporation of meta-structure information substantially improves performance of an unsupervised segmentation model developed for binary semantic segmentation. We define meta-structures mathematically as spatial density distributions and show both theoretically and experimentally how this formulation explains key observed learning behavior of DNNs.

Lunjun Zhang, Ge Yang, Bradly C. Stadie

Planning - the ability to analyze the structure of a problem in the large and decompose it into interrelated subproblems - is a hallmark of human intelligence. While deep reinforcement learning (RL) has shown great promise for solving relatively straightforward control tasks, it remains an open problem how to best incorporate planning into existing deep RL paradigms to handle increasingly complex environments. One prominent framework, Model-Based RL, learns a world model and plans using step-by-step virtual rollouts. This type of world model quickly diverges from reality when the planning horizon increases, thus struggling at long-horizon planning. How can we learn world models that endow agents with the ability to do temporally extended reasoning? In this work, we propose to learn graph-structured world models composed of sparse, multi-step transitions. We devise a novel algorithm to learn latent landmarks that are scattered (in terms of reachability) across the goal space as the nodes on the graph. In this same graph, the edges are the reachability estimates distilled from Q-functions. On a variety of high-dimensional continuous control tasks ranging from robotic manipulation to navigation, we demonstrate that our method, named L3P, significantly outperforms prior work, and is oftentimes the only method capable of leveraging both the robustness of model-free RL and generalization of graph-search algorithms. We believe our work is an important step towards scalable planning in reinforcement learning.

Liangyong Yu, Ran Li, Xiangrui Zeng et al.

Motivation: Cryo-Electron Tomography (cryo-ET) visualizes structure and spatial organization of macromolecules and their interactions with other subcellular components inside single cells in the close-to-native state at sub-molecular resolution. Such information is critical for the accurate understanding of cellular processes. However, subtomogram classification remains one of the major challenges for the systematic recognition and recovery of the macromolecule structures in cryo-ET because of imaging limits and data quantity. Recently, deep learning has significantly improved the throughput and accuracy of large-scale subtomogram classification. However often it is difficult to get enough high-quality annotated subtomogram data for supervised training due to the enormous expense of labeling. To tackle this problem, it is beneficial to utilize another already annotated dataset to assist the training process. However, due to the discrepancy of image intensity distribution between source domain and target domain, the model trained on subtomograms in source domainmay perform poorly in predicting subtomogram classes in the target domain. Results: In this paper, we adapt a few shot domain adaptation method for deep learning based cross-domain subtomogram classification. The essential idea of our method consists of two parts: 1) take full advantage of the distribution of plentiful unlabeled target domain data, and 2) exploit the correlation between the whole source domain dataset and few labeled target domain data. Experiments conducted on simulated and real datasets show that our method achieves significant improvement on cross domain subtomogram classification compared with baseline methods.

Ge Yang, Amy Zhang, Ari S. Morcos et al.

In this paper we introduce plan2vec, an unsupervised representation learning approach that is inspired by reinforcement learning. Plan2vec constructs a weighted graph on an image dataset using near-neighbor distances, and then extrapolates this local metric to a global embedding by distilling path-integral over planned path. When applied to control, plan2vec offers a way to learn goal-conditioned value estimates that are accurate over long horizons that is both compute and sample efficient. We demonstrate the effectiveness of plan2vec on one simulated and two challenging real-world image datasets. Experimental results show that plan2vec successfully amortizes the planning cost, enabling reactive planning that is linear in memory and computation complexity rather than exhaustive over the entire state space.

Yuanhao Guo, Fons J. Verbeek, Ge Yang

Robust and accurate camera calibration is essential for 3D reconstruction in light microscopy under circular motion. Conventional methods require either accurate key point matching or precise segmentation of the axial-view images. Both remain challenging because specimens often exhibit transparency/translucency in a light microscope. To address those issues, we propose a probabilistic inference based method for the camera calibration that does not require sophisticated image pre-processing. Based on 3D projective geometry, our method assigns a probability on each of a range of voxels that cover the whole object. The probability indicates the likelihood of a voxel belonging to the object to be reconstructed. Our method maximizes a joint probability that distinguishes the object from the background. Experimental results show that the proposed method can accurately recover camera configurations in both light microscopy and natural scene imaging. Furthermore, the method can be used to produce high-fidelity 3D reconstructions and accurate 3D measurements.

Thanard Kurutach, Aviv Tamar, Ge Yang et al.

In recent years, deep generative models have been shown to 'imagine' convincing high-dimensional observations such as images, audio, and even video, learning directly from raw data. In this work, we ask how to imagine goal-directed visual plans -- a plausible sequence of observations that transition a dynamical system from its current configuration to a desired goal state, which can later be used as a reference trajectory for control. We focus on systems with high-dimensional observations, such as images, and propose an approach that naturally combines representation learning and planning. Our framework learns a generative model of sequential observations, where the generative process is induced by a transition in a low-dimensional planning model, and an additional noise. By maximizing the mutual information between the generated observations and the transition in the planning model, we obtain a low-dimensional representation that best explains the causal nature of the data. We structure the planning model to be compatible with efficient planning algorithms, and we propose several such models based on either discrete or continuous states. Finally, to generate a visual plan, we project the current and goal observations onto their respective states in the planning model, plan a trajectory, and then use the generative model to transform the trajectory to a sequence of observations. We demonstrate our method on imagining plausible visual plans of rope manipulation.

Bradly C. Stadie, Ge Yang, Rein Houthooft et al.

We consider the problem of exploration in meta reinforcement learning. Two new meta reinforcement learning algorithms are suggested: E-MAML and E-$\text{RL}^2$. Results are presented on a novel environment we call `Krazy World' and a set of maze environments. We show E-MAML and E-$\text{RL}^2$ deliver better performance on tasks where exploration is important.