Tiejun Huang

Yuxin Fang, Wen Wang, Binhui Xie et al. · meta-ai

We launch EVA, a vision-centric foundation model to explore the limits of visual representation at scale using only publicly accessible data. EVA is a vanilla ViT pre-trained to reconstruct the masked out image-text aligned vision features conditioned on visible image patches. Via this pretext task, we can efficiently scale up EVA to one billion parameters, and sets new records on a broad range of representative vision downstream tasks, such as image recognition, video action recognition, object detection, instance segmentation and semantic segmentation without heavy supervised training. Moreover, we observe quantitative changes in scaling EVA result in qualitative changes in transfer learning performance that are not present in other models. For instance, EVA takes a great leap in the challenging large vocabulary instance segmentation task: our model achieves almost the same state-of-the-art performance on LVISv1.0 dataset with over a thousand categories and COCO dataset with only eighty categories. Beyond a pure vision encoder, EVA can also serve as a vision-centric, multi-modal pivot to connect images and text. We find initializing the vision tower of a giant CLIP from EVA can greatly stabilize the training and outperform the training from scratch counterpart with much fewer samples and less compute, providing a new direction for scaling up and accelerating the costly training of multi-modal foundation models. To facilitate future research, we release all the code and models at https://github.com/baaivision/EVA.

Yuxin Fang, Quan Sun, Xinggang Wang et al.

We launch EVA-02, a next-generation Transformer-based visual representation pre-trained to reconstruct strong and robust language-aligned vision features via masked image modeling. With an updated plain Transformer architecture as well as extensive pre-training from an open & accessible giant CLIP vision encoder, EVA-02 demonstrates superior performance compared to prior state-of-the-art approaches across various representative vision tasks, while utilizing significantly fewer parameters and compute budgets. Notably, using exclusively publicly accessible training data, EVA-02 with only 304M parameters achieves a phenomenal 90.0 fine-tuning top-1 accuracy on ImageNet-1K val set. Additionally, our EVA-02-CLIP can reach up to 80.4 zero-shot top-1 on ImageNet-1K, outperforming the previous largest & best open-sourced CLIP with only ~1/6 parameters and ~1/6 image-text training data. We offer four EVA-02 variants in various model sizes, ranging from 6M to 304M parameters, all with impressive performance. To facilitate open access and open research, we release the complete suite of EVA-02 to the community at https://github.com/baaivision/EVA/tree/master/EVA-02.

Xinlong Wang, Xiaosong Zhang, Zhengxiong Luo et al. · tsinghua

While next-token prediction is considered a promising path towards artificial general intelligence, it has struggled to excel in multimodal tasks, which are still dominated by diffusion models (e.g., Stable Diffusion) and compositional approaches (e.g., CLIP combined with LLMs). In this paper, we introduce Emu3, a new suite of state-of-the-art multimodal models trained solely with next-token prediction. By tokenizing images, text, and videos into a discrete space, we train a single transformer from scratch on a mixture of multimodal sequences. Emu3 outperforms several well-established task-specific models in both generation and perception tasks, surpassing flagship models such as SDXL and LLaVA-1.6, while eliminating the need for diffusion or compositional architectures. Emu3 is also capable of generating high-fidelity video via predicting the next token in a video sequence. We simplify complex multimodal model designs by converging on a singular focus: tokens, unlocking great potential for scaling both during training and inference. Our results demonstrate that next-token prediction is a promising path towards building general multimodal intelligence beyond language. We open-source key techniques and models to support further research in this direction.

Zecheng Hao, Tong Bu, Jianhao Ding et al. · pku

Spiking Neural Networks (SNNs) have received extensive academic attention due to the unique properties of low power consumption and high-speed computing on neuromorphic chips. Among various training methods of SNNs, ANN-SNN conversion has shown the equivalent level of performance as ANNs on large-scale datasets. However, unevenness error, which refers to the deviation caused by different temporal sequences of spike arrival on activation layers, has not been effectively resolved and seriously suffers the performance of SNNs under the condition of short time-steps. In this paper, we make a detailed analysis of unevenness error and divide it into four categories. We point out that the case of the ANN output being zero while the SNN output being larger than zero accounts for the largest percentage. Based on this, we theoretically prove the sufficient and necessary conditions of this case and propose an optimization strategy based on residual membrane potential to reduce unevenness error. The experimental results show that the proposed method achieves state-of-the-art performance on CIFAR-10, CIFAR-100, and ImageNet datasets. For example, we reach top-1 accuracy of 64.32\% on ImageNet with 10-steps. To the best of our knowledge, this is the first time ANN-SNN conversion can simultaneously achieve high accuracy and ultra-low-latency on the complex dataset. Code is available at https://github.com/hzc1208/ANN2SNN\_SRP.

Zecheng Hao, Jianhao Ding, Tong Bu et al. · pku

Spiking Neural Networks (SNNs) have attracted great attention due to their distinctive characteristics of low power consumption and temporal information processing. ANN-SNN conversion, as the most commonly used training method for applying SNNs, can ensure that converted SNNs achieve comparable performance to ANNs on large-scale datasets. However, the performance degrades severely under low quantities of time-steps, which hampers the practical applications of SNNs to neuromorphic chips. In this paper, instead of evaluating different conversion errors and then eliminating these errors, we define an offset spike to measure the degree of deviation between actual and desired SNN firing rates. We perform a detailed analysis of offset spike and note that the firing of one additional (or one less) spike is the main cause of conversion errors. Based on this, we propose an optimization strategy based on shifting the initial membrane potential and we theoretically prove the corresponding optimal shifting distance for calibrating the spike. In addition, we also note that our method has a unique iterative property that enables further reduction of conversion errors. The experimental results show that our proposed method achieves state-of-the-art performance on CIFAR-10, CIFAR-100, and ImageNet datasets. For example, we reach a top-1 accuracy of 67.12% on ImageNet when using 6 time-steps. To the best of our knowledge, this is the first time an ANN-SNN conversion has been shown to simultaneously achieve high accuracy and ultralow latency on complex datasets. Code is available at https://github.com/hzc1208/ANN2SNN_COS.

Bo Zhao, Boya Wu, Muyang He et al.

Thanks to the emerging of foundation models, the large language and vision models are integrated to acquire the multimodal ability of visual captioning, question answering, etc. Although existing multimodal models present impressive performance of visual understanding and reasoning, their limits are still largely under-explored due to the scarcity of high-quality instruction tuning data. To push the limits of multimodal capability, we Scale up Visual Instruction Tuning (SVIT) by constructing a dataset of 4.2 million visual instruction tuning data including 1.6M conversation question-answer (QA) pairs, 1.6M complex reasoning QA pairs, 1.0M referring QA pairs and 106K detailed image descriptions. Besides the volume, the proposed dataset is also featured by the high quality and rich diversity, which is generated by prompting GPT-4 with the abundant manual annotations of images. We also propose a new data recipe to select subset with better diversity and balance, which evokes model's superior capabilities. Extensive experiments verify that SVIT-v1.5, trained on the proposed dataset, outperforms state-of-the-art Multimodal Large Language Models on popular benchmarks. The data and code are publicly available at https://github.com/BAAI-DCAI/Visual-Instruction-Tuning.

Yajing Zheng, Jiyuan Zhang, Rui Zhao et al. · pku

SpikeCV is a new open-source computer vision platform for the spike camera, which is a neuromorphic visual sensor that has developed rapidly in recent years. In the spike camera, each pixel position directly accumulates the light intensity and asynchronously fires spikes. The output binary spikes can reach a frequency of 40,000 Hz. As a new type of visual expression, spike sequence has high spatiotemporal completeness and preserves the continuous visual information of the external world. Taking advantage of the low latency and high dynamic range of the spike camera, many spike-based algorithms have made significant progress, such as high-quality imaging and ultra-high-speed target detection. To build up a community ecology for the spike vision to facilitate more users to take advantage of the spike camera, SpikeCV provides a variety of ultra-high-speed scene datasets, hardware interfaces, and an easy-to-use modules library. SpikeCV focuses on encapsulation for spike data, standardization for dataset interfaces, modularization for vision tasks, and real-time applications for challenging scenes. With the advent of the open-source Python ecosystem, modules of SpikeCV can be used as a Python library to fulfilled most of the numerical analysis needs of researchers. We demonstrate the efficiency of the SpikeCV on offline inference and real-time applications. The project repository address are \url{https://openi.pcl.ac.cn/Cordium/SpikeCV} and \url{https://github.com/Zyj061/SpikeCV

Muyang He, Shuo Yang, Tiejun Huang et al.

The state of the art of many learning tasks, e.g., image classification, is advanced by collecting larger datasets and then training larger models on them. As the outcome, the increasing computational cost is becoming unaffordable. In this paper, we investigate how to prune the large-scale datasets, and thus produce an informative subset for training sophisticated deep models with negligible performance drop. We propose a simple yet effective dataset pruning method by exploring both the prediction uncertainty and training dynamics. We study dataset pruning by measuring the variation of predictions during the whole training process on large-scale datasets, i.e., ImageNet-1K and ImageNet-21K, and advanced models, i.e., Swin Transformer and ConvNeXt. Extensive experimental results indicate that our method outperforms the state of the art and achieves 25% lossless pruning ratio on both ImageNet-1K and ImageNet-21K. The code and pruned datasets are available at https://github.com/BAAI-DCAI/Dataset-Pruning.

Shitao Xiao, Yueze Wang, Junjie Zhou et al.

The emergence of Large Language Models (LLMs) has unified language generation tasks and revolutionized human-machine interaction. However, in the realm of image generation, a unified model capable of handling various tasks within a single framework remains largely unexplored. In this work, we introduce OmniGen, a new diffusion model for unified image generation. OmniGen is characterized by the following features: 1) Unification: OmniGen not only demonstrates text-to-image generation capabilities but also inherently supports various downstream tasks, such as image editing, subject-driven generation, and visual-conditional generation. 2) Simplicity: The architecture of OmniGen is highly simplified, eliminating the need for additional plugins. Moreover, compared to existing diffusion models, it is more user-friendly and can complete complex tasks end-to-end through instructions without the need for extra intermediate steps, greatly simplifying the image generation workflow. 3) Knowledge Transfer: Benefit from learning in a unified format, OmniGen effectively transfers knowledge across different tasks, manages unseen tasks and domains, and exhibits novel capabilities. We also explore the model's reasoning capabilities and potential applications of the chain-of-thought mechanism. This work represents the first attempt at a general-purpose image generation model, and we will release our resources at https://github.com/VectorSpaceLab/OmniGen to foster future advancements.

Quan Sun, Qiying Yu, Yufeng Cui et al. · tsinghua

We present Emu, a Transformer-based multimodal foundation model, which can seamlessly generate images and texts in multimodal context. This omnivore model can take in any single-modality or multimodal data input indiscriminately (e.g., interleaved image, text and video) through a one-model-for-all autoregressive training process. First, visual signals are encoded into embeddings, and together with text tokens form an interleaved input sequence. Emu is then end-to-end trained with a unified objective of classifying the next text token or regressing the next visual embedding in the multimodal sequence. This versatile multimodality empowers the exploration of diverse pretraining data sources at scale, such as videos with interleaved frames and text, webpages with interleaved images and text, as well as web-scale image-text pairs and video-text pairs. Emu can serve as a generalist multimodal interface for both image-to-text and text-to-image tasks, and supports in-context image and text generation. Across a broad range of zero-shot/few-shot tasks including image captioning, visual question answering, video question answering and text-to-image generation, Emu demonstrates superb performance compared to state-of-the-art large multimodal models. Extended capabilities such as multimodal assistants via instruction tuning are also demonstrated with impressive performance.

Xinlong Wang, Wen Wang, Yue Cao et al.

In-context learning, as a new paradigm in NLP, allows the model to rapidly adapt to various tasks with only a handful of prompts and examples. But in computer vision, the difficulties for in-context learning lie in that tasks vary significantly in the output representations, thus it is unclear how to define the general-purpose task prompts that the vision model can understand and transfer to out-of-domain tasks. In this work, we present Painter, a generalist model which addresses these obstacles with an "image"-centric solution, that is, to redefine the output of core vision tasks as images, and specify task prompts as also images. With this idea, our training process is extremely simple, which performs standard masked image modeling on the stitch of input and output image pairs. This makes the model capable of performing tasks conditioned on visible image patches. Thus, during inference, we can adopt a pair of input and output images from the same task as the input condition, to indicate which task to perform. Without bells and whistles, our generalist Painter can achieve competitive performance compared to well-established task-specific models, on seven representative vision tasks ranging from high-level visual understanding to low-level image processing. In addition, Painter significantly outperforms recent generalist models on several challenging tasks.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large Language Models (LLMs) represent a significant stride toward Artificial General Intelligence. As scaling laws underscore the potential of increasing model sizes, the academic community has intensified its investigations into LLMs with capacities exceeding 50 billion parameters. This technical report builds on our prior work with Tele-FLM (also known as FLM-2), a publicly available 52-billion-parameter model. We delve into two primary areas: we first discuss our observation of Supervised Fine-tuning (SFT) on Tele-FLM-52B, which supports the "less is more" approach for SFT data construction; second, we demonstrate our experiments and analyses on the best practices for progressively growing a model from 52 billion to 102 billion, and subsequently to 1 trillion parameters. We will open-source a 1T model checkpoint, namely Tele-FLM-1T, to advance further training and research.

Xinlong Wang, Xiaosong Zhang, Yue Cao et al.

We present SegGPT, a generalist model for segmenting everything in context. We unify various segmentation tasks into a generalist in-context learning framework that accommodates different kinds of segmentation data by transforming them into the same format of images. The training of SegGPT is formulated as an in-context coloring problem with random color mapping for each data sample. The objective is to accomplish diverse tasks according to the context, rather than relying on specific colors. After training, SegGPT can perform arbitrary segmentation tasks in images or videos via in-context inference, such as object instance, stuff, part, contour, and text. SegGPT is evaluated on a broad range of tasks, including few-shot semantic segmentation, video object segmentation, semantic segmentation, and panoptic segmentation. Our results show strong capabilities in segmenting in-domain and out-of-domain targets, either qualitatively or quantitatively.

Chuang Zhu, Kebin Liu, Wenqi Tang et al.

The divergence between labeled training data and unlabeled testing data is a significant challenge for recent deep learning models. Unsupervised domain adaptation (UDA) attempts to solve such problem. Recent works show that self-training is a powerful approach to UDA. However, existing methods have difficulty in balancing the scalability and performance. In this paper, we propose a hard-aware instance adaptive self-training framework for UDA on the task of semantic segmentation. To effectively improve the quality and diversity of pseudo-labels, we develop a novel pseudo-label generation strategy with an instance adaptive selector. We further enrich the hard class pseudo-labels with inter-image information through a skillfully designed hard-aware pseudo-label augmentation. Besides, we propose the region-adaptive regularization to smooth the pseudo-label region and sharpen the non-pseudo-label region. For the non-pseudo-label region, consistency constraint is also constructed to introduce stronger supervision signals during model optimization. Our method is so concise and efficient that it is easy to be generalized to other UDA methods. Experiments on GTA5 to Cityscapes, SYNTHIA to Cityscapes, and Cityscapes to Oxford RobotCar demonstrate the superior performance of our approach compared with the state-of-the-art methods. Our codes are available at https://github.com/bupt-ai-cz/HIAST.

Jiaming Liu, Qizhe Zhang, Xiaoqi Li et al.

Neuromorphic spike data, an upcoming modality with high temporal resolution, has shown promising potential in autonomous driving by mitigating the challenges posed by high-velocity motion blur. However, training the spike depth estimation network holds significant challenges in two aspects: sparse spatial information for pixel-wise tasks and difficulties in achieving paired depth labels for temporally intensive spike streams. Therefore, we introduce open-source RGB data to support spike depth estimation, leveraging its annotations and spatial information. The inherent differences in modalities and data distribution make it challenging to directly apply transfer learning from open-source RGB to target spike data. To this end, we propose a cross-modality cross-domain (BiCross) framework to realize unsupervised spike depth estimation by introducing simulated mediate source spike data. Specifically, we design a Coarse-to-Fine Knowledge Distillation (CFKD) approach to facilitate comprehensive cross-modality knowledge transfer while preserving the unique strengths of both modalities, utilizing a spike-oriented uncertainty scheme. Then, we propose a Self-Correcting Teacher-Student (SCTS) mechanism to screen out reliable pixel-wise pseudo labels and ease the domain shift of the student model, which avoids error accumulation in target spike data. To verify the effectiveness of BiCross, we conduct extensive experiments on four scenarios, including Synthetic to Real, Extreme Weather, Scene Changing, and Real Spike. Our method achieves state-of-the-art (SOTA) performances, compared with RGB-oriented unsupervised depth estimation methods. Code and dataset: https://github.com/Theia-4869/BiCross

Yuxin Du, Fan Bai, Tiejun Huang et al.

Precise image segmentation provides clinical study with instructive information. Despite the remarkable progress achieved in medical image segmentation, there is still an absence of a 3D foundation segmentation model that can segment a wide range of anatomical categories with easy user interaction. In this paper, we propose a 3D foundation segmentation model, named SegVol, supporting universal and interactive volumetric medical image segmentation. By scaling up training data to 90K unlabeled Computed Tomography (CT) volumes and 6K labeled CT volumes, this foundation model supports the segmentation of over 200 anatomical categories using semantic and spatial prompts. To facilitate efficient and precise inference on volumetric images, we design a zoom-out-zoom-in mechanism. Extensive experiments on 22 anatomical segmentation tasks verify that SegVol outperforms the competitors in 19 tasks, with improvements up to 37.24% compared to the runner-up methods. We demonstrate the effectiveness and importance of specific designs by ablation study. We expect this foundation model can promote the development of volumetric medical image analysis. The model and code are publicly available at: https://github.com/BAAI-DCAI/SegVol.

Shihan Wu, Xuecheng Liu, Shaoxuan Xie et al.

Despite the critical role of bimanual manipulation in endowing robots with human-like dexterity, large-scale and diverse datasets remain scarce due to the significant hardware heterogeneity across bimanual robotic platforms. To bridge this gap, we introduce RoboCOIN, a large-scale multi-embodiment bimanual manipulation dataset comprising over 180,000 demonstrations collected from 15 distinct robotic platforms. Spanning 16 diverse environments-including residential, commercial, and industrial settings-the dataset features 421 bimanual tasks systematically categorized by 39 bimanual collaboration actions and 432 objects. A key innovation of our work is the hierarchical capability pyramid, which provides granular annotations ranging from trajectory-level concepts to segment-level subtasks and frame-level kinematics. Furthermore, we present CoRobot, an efficient data processing pipeline powered by the Robot Trajectory Markup Language (RTML), designed to facilitate quality assessment, automated annotation, and unified multi-embodiment and data management. Extensive experiments demonstrate the effectiveness of RoboCOIN in enhancing the performance of various bimanual manipulation models across a wide spectrum of robotic embodiments. The entire dataset and codebase are fully open-sourced, providing a valuable resource for advancing research in bimanual and multi-embodiment manipulation.

Lujie Xia, Ziluo Ding, Rui Zhao et al. · pku

Efficiently selecting an appropriate spike stream data length to extract precise information is the key to the spike vision tasks. To address this issue, we propose a dynamic timing representation for spike streams. Based on multi-layers architecture, it applies dilated convolutions on temporal dimension to extract features on multi-temporal scales with few parameters. And we design layer attention to dynamically fuse these features. Moreover, we propose an unsupervised learning method for optical flow estimation in a spike-based manner to break the dependence on labeled data. In addition, to verify the robustness, we also build a spike-based synthetic validation dataset for extreme scenarios in autonomous driving, denoted as SSES dataset. It consists of various corner cases. Experiments show that our method can predict optical flow from spike streams in different high-speed scenes, including real scenes. For instance, our method gets $15\%$ and $19\%$ error reduction from the best spike-based work, SCFlow, in $Δt=10$ and $Δt=20$ respectively which are the same settings as the previous works.

Xian Zhong, Shengwang Hu, Wenxuan Liu et al.

Spiking neural networks (SNNs) have garnered significant attention for their low power consumption and high biological interpretability. Their rich spatio-temporal information processing capability and event-driven nature make them ideally well-suited for neuromorphic datasets. However, current SNNs struggle to balance accuracy and latency in classifying these datasets. In this paper, we propose Hybrid Step-wise Distillation (HSD) method, tailored for neuromorphic datasets, to mitigate the notable decline in performance at lower time steps. Our work disentangles the dependency between the number of event frames and the time steps of SNNs, utilizing more event frames during the training stage to improve performance, while using fewer event frames during the inference stage to reduce latency. Nevertheless, the average output of SNNs across all time steps is susceptible to individual time step with abnormal outputs, particularly at extremely low time steps. To tackle this issue, we implement Step-wise Knowledge Distillation (SKD) module that considers variations in the output distribution of SNNs at each time step. Empirical evidence demonstrates that our method yields competitive performance in classification tasks on neuromorphic datasets, especially at lower time steps. Our code will be available at: {https://github.com/hsw0929/HSD}.

Ziluo Ding, Wanpeng Zhang, Junpeng Yue et al. · pku

We investigate the use of natural language to drive the generalization of policies in multi-agent settings. Unlike single-agent settings, the generalization of policies should also consider the influence of other agents. Besides, with the increasing number of entities in multi-agent settings, more agent-entity interactions are needed for language grounding, and the enormous search space could impede the learning process. Moreover, given a simple general instruction,e.g., beating all enemies, agents are required to decompose it into multiple subgoals and figure out the right one to focus on. Inspired by previous work, we try to address these issues at the entity level and propose a novel framework for language grounding in multi-agent reinforcement learning, entity divider (EnDi). EnDi enables agents to independently learn subgoal division at the entity level and act in the environment based on the associated entities. The subgoal division is regularized by opponent modeling to avoid subgoal conflicts and promote coordinated strategies. Empirically, EnDi demonstrates the strong generalization ability to unseen games with new dynamics and expresses the superiority over existing methods.

Liwen hu, Lei Ma, Zhaofei Yu et al. · pku

As a neuromorphic sensor with high temporal resolution, the spike camera shows enormous potential in high-speed visual tasks. However, the high-speed sampling of light propagation processes by existing cameras brings unavoidable noise phenomena. Eliminating the unique noise in spike stream is always a key point for spike-based methods. No previous work has addressed the detailed noise mechanism of the spike camera. To this end, we propose a systematic noise model for spike camera based on its unique circuit. In addition, we carefully constructed the noise evaluation equation and experimental scenarios to measure noise variables. Based on our noise model, the first benchmark for spike stream denoising is proposed which includes clear (noisy) spike stream. Further, we design a tailored spike stream denoising framework (DnSS) where denoised spike stream is obtained by decoding inferred inter-spike intervals. Experiments show that DnSS has promising performance on the proposed benchmark. Eventually, DnSS can be generalized well on real spike stream.

Jianing Li, Jiaming Liu, Xiaobao Wei et al. · pku

Depth estimation is essential for various important real-world applications such as autonomous driving. However, it suffers from severe performance degradation in high-velocity scenario since traditional cameras can only capture blurred images. To deal with this problem, the spike camera is designed to capture the pixel-wise luminance intensity at high frame rate. However, depth estimation with spike camera remains very challenging using traditional monocular or stereo depth estimation algorithms, which are based on the photometric consistency. In this paper, we propose a novel Uncertainty-Guided Depth Fusion (UGDF) framework to fuse the predictions of monocular and stereo depth estimation networks for spike camera. Our framework is motivated by the fact that stereo spike depth estimation achieves better results at close range while monocular spike depth estimation obtains better results at long range. Therefore, we introduce a dual-task depth estimation architecture with a joint training strategy and estimate the distributed uncertainty to fuse the monocular and stereo results. In order to demonstrate the advantage of spike depth estimation over traditional camera depth estimation, we contribute a spike-depth dataset named CitySpike20K, which contains 20K paired samples, for spike depth estimation. UGDF achieves state-of-the-art results on CitySpike20K, surpassing all monocular or stereo spike depth estimation baselines. We conduct extensive experiments to evaluate the effectiveness and generalization of our method on CitySpike20K. To the best of our knowledge, our framework is the first dual-task fusion framework for spike camera depth estimation. Code and dataset will be released.

Yan Shu, Zheng Liu, Peitian Zhang et al.

Long video understanding poses a significant challenge for current Multi-modal Large Language Models (MLLMs). Notably, the MLLMs are constrained by their limited context lengths and the substantial costs while processing long videos. Although several existing methods attempt to reduce visual tokens, their strategies encounter severe bottleneck, restricting MLLMs' ability to perceive fine-grained visual details. In this work, we propose Video-XL, a novel approach that leverages MLLMs' inherent key-value (KV) sparsification capacity to condense the visual input. Specifically, we introduce a new special token, the Visual Summarization Token (VST), for each interval of the video, which summarizes the visual information within the interval as its associated KV. The VST module is trained by instruction fine-tuning, where two optimizing strategies are offered. 1.Curriculum learning, where VST learns to make small (easy) and large compression (hard) progressively. 2. Composite data curation, which integrates single-image, multi-image, and synthetic data to overcome the scarcity of long-video instruction data. The compression quality is further improved by dynamic compression, which customizes compression granularity based on the information density of different video intervals. Video-XL's effectiveness is verified from three aspects. First, it achieves a superior long-video understanding capability, outperforming state-of-the-art models of comparable sizes across multiple popular benchmarks. Second, it effectively preserves video information, with minimal compression loss even at 16x compression ratio. Third, it realizes outstanding cost-effectiveness, enabling high-quality processing of thousands of frames on a single A100 GPU.

Yu Wang, Xuelin Qian, Jingyang Huo et al.

We present a significant breakthrough in 3D shape generation by scaling it to unprecedented dimensions. Through the adaptation of the Auto-Regressive model and the utilization of large language models, we have developed a remarkable model with an astounding 3.6 billion trainable parameters, establishing it as the largest 3D shape generation model to date, named Argus-3D. Our approach addresses the limitations of existing methods by enhancing the quality and diversity of generated 3D shapes. To tackle the challenges of high-resolution 3D shape generation, our model incorporates tri-plane features as latent representations, effectively reducing computational complexity. Additionally, we introduce a discrete codebook for efficient quantization of these representations. Leveraging the power of transformers, we enable multi-modal conditional generation, facilitating the production of diverse and visually impressive 3D shapes. To train our expansive model, we leverage an ensemble of publicly-available 3D datasets, consisting of a comprehensive collection of approximately 900,000 objects from renowned repositories such as ModelNet40, ShapeNet, Pix3D, 3D-Future, and Objaverse. This diverse dataset empowers our model to learn from a wide range of object variations, bolstering its ability to generate high-quality and diverse 3D shapes. Extensive experimentation demonstrate the remarkable efficacy of our approach in significantly improving the visual quality of generated 3D shapes. By pushing the boundaries of 3D generation, introducing novel methods for latent representation learning, and harnessing the power of transformers for multi-modal conditional generation, our contributions pave the way for substantial advancements in the field. Our work unlocks new possibilities for applications in gaming, virtual reality, product design, and other domains that demand high-quality and diverse 3D objects.

Junsheng Zhou, Jinsheng Wang, Baorui Ma et al.

Scaling up representations for images or text has been extensively investigated in the past few years and has led to revolutions in learning vision and language. However, scalable representation for 3D objects and scenes is relatively unexplored. In this work, we present Uni3D, a 3D foundation model to explore the unified 3D representation at scale. Uni3D uses a 2D initialized ViT end-to-end pretrained to align the 3D point cloud features with the image-text aligned features. Via the simple architecture and pretext task, Uni3D can leverage abundant 2D pretrained models as initialization and image-text aligned models as the target, unlocking the great potential of 2D models and scaling-up strategies to the 3D world. We efficiently scale up Uni3D to one billion parameters, and set new records on a broad range of 3D tasks, such as zero-shot classification, few-shot classification, open-world understanding and part segmentation. We show that the strong Uni3D representation also enables applications such as 3D painting and retrieval in the wild. We believe that Uni3D provides a new direction for exploring both scaling up and efficiency of the representation in 3D domain.

Hongjin Qian, Zheng Liu, Peitian Zhang et al.

Processing long contexts presents a significant challenge for large language models (LLMs). While recent advancements allow LLMs to handle much longer contexts than before (e.g., 32K or 128K tokens), it is computationally expensive and can still be insufficient for many applications. Retrieval-Augmented Generation (RAG) is considered a promising strategy to address this problem. However, conventional RAG methods face inherent limitations because of two underlying requirements: 1) explicitly stated queries, and 2) well-structured knowledge. These conditions, however, do not hold in general long-context processing tasks. In this work, we propose MemoRAG, a novel RAG framework empowered by global memory-augmented retrieval. MemoRAG features a dual-system architecture. First, it employs a light but long-range system to create a global memory of the long context. Once a task is presented, it generates draft answers, providing useful clues for the retrieval tools to locate relevant information within the long context. Second, it leverages an expensive but expressive system, which generates the final answer based on the retrieved information. Building upon this fundamental framework, we realize the memory module in the form of KV compression, and reinforce its memorization and cluing capacity from the Generation quality's Feedback (a.k.a. RLGF). In our experiments, MemoRAG achieves superior performances across a variety of long-context evaluation tasks, not only complex scenarios where traditional RAG methods struggle, but also simpler ones where RAG is typically applied.

Xijie Xiang, Lin Zhu, Jianing Li et al.

The event camera is a novel bio-inspired vision sensor. When the brightness change exceeds the preset threshold, the sensor generates events asynchronously. The number of valid events directly affects the performance of event-based tasks, such as reconstruction, detection, and recognition. However, when in low-brightness or slow-moving scenes, events are often sparse and accompanied by noise, which poses challenges for event-based tasks. To solve these challenges, we propose an event temporal up-sampling algorithm1 to generate more effective and reliable events. The main idea of our algorithm is to generate up-sampling events on the event motion trajectory. First, we estimate the event motion trajectory by contrast maximization algorithm and then up-sampling the events by temporal point processes. Experimental results show that up-sampling events can provide more effective information and improve the performance of downstream tasks, such as improving the quality of reconstructed images and increasing the accuracy of object detection.

Yufeng Cui, Honghao Chen, Haoge Deng et al.

We introduce Emu3.5, a large-scale multimodal world model that natively predicts the next state across vision and language. Emu3.5 is pre-trained end-to-end with a unified next-token prediction objective on a corpus of vision-language interleaved data containing over 10 trillion tokens, primarily derived from sequential frames and transcripts of internet videos. The model naturally accepts interleaved vision-language inputs and generates interleaved vision-language outputs. Emu3.5 is further post-trained with large-scale reinforcement learning to enhance multimodal reasoning and generation. To improve inference efficiency, we propose Discrete Diffusion Adaptation (DiDA), which converts token-by-token decoding into bidirectional parallel prediction, accelerating per-image inference by about 20x without sacrificing performance. Emu3.5 exhibits strong native multimodal capabilities, including long-horizon vision-language generation, any-to-image (X2I) generation, and complex text-rich image generation. It also exhibits generalizable world-modeling abilities, enabling spatiotemporally consistent world exploration and open-world embodied manipulation across diverse scenarios and tasks. For comparison, Emu3.5 achieves performance comparable to Gemini 2.5 Flash Image (Nano Banana) on image generation and editing tasks and demonstrates superior results on a suite of interleaved generation tasks. We open-source Emu3.5 at https://github.com/baaivision/Emu3.5 to support community research.

Xintao Zong, Xian Zhong, Wenxuan Liu et al.

Spiking neural networks (SNNs) have recently shown strong potential in unimodal visual and textual tasks, yet building a directly trained, low-energy, and high-performance SNN for multimodal applications such as image-text retrieval (ITR) remains highly challenging. Existing artificial neural network (ANN)-based methods often pursue richer unimodal semantics using deeper and more complex architectures, while overlooking cross-modal interaction, retrieval latency, and energy efficiency. To address these limitations, we present a brain-inspired Cross-Modal Spike Fusion network (CMSF) and apply it to ITR for the first time. The proposed spike fusion mechanism integrates unimodal features at the spike level, generating enhanced multimodal representations that act as soft supervisory signals to refine unimodal spike embeddings, effectively mitigating semantic loss within CMSF. Despite requiring only two time steps, CMSF achieves top-tier retrieval accuracy, surpassing state-of-the-art ANN counterparts while maintaining exceptionally low energy consumption and high retrieval speed. This work marks a significant step toward multimodal SNNs, offering a brain-inspired framework that unifies temporal dynamics with cross-modal alignment and provides new insights for future spiking-based multimodal research. The code is available at https://github.com/zxt6174/CMSF.

Shiyan Chen, Jiyuan Zhang, Zhaofei Yu et al.

Self-supervised denoising has attracted widespread attention due to its ability to train without clean images. However, noise in real-world scenarios is often spatially correlated, which causes many self-supervised algorithms that assume pixel-wise independent noise to perform poorly. Recent works have attempted to break noise correlation with downsampling or neighborhood masking. However, denoising on downsampled subgraphs can lead to aliasing effects and loss of details due to a lower sampling rate. Furthermore, the neighborhood masking methods either come with high computational complexity or do not consider local spatial preservation during inference. Through the analysis of existing methods, we point out that the key to obtaining high-quality and texture-rich results in real-world self-supervised denoising tasks is to train at the original input resolution structure and use asymmetric operations during training and inference. Based on this, we propose Asymmetric Tunable Blind-Spot Network (AT-BSN), where the blind-spot size can be freely adjusted, thus better balancing noise correlation suppression and image local spatial destruction during training and inference. In addition, we regard the pre-trained AT-BSN as a meta-teacher network capable of generating various teacher networks by sampling different blind-spots. We propose a blind-spot based multi-teacher distillation strategy to distill a lightweight network, significantly improving performance. Experimental results on multiple datasets prove that our method achieves state-of-the-art, and is superior to other self-supervised algorithms in terms of computational overhead and visual effects.

Yang Yue, Shenjian Zhang, Yonghong Tian et al.

Deep Neural Networks (DNNs) are vulnerable to elaborately designed adversarial noise, although they have achieved extraordinary success in many tasks. Compared with DNNs, the human visual system is highly robust. However, it is unclear how the human visual system defends against adversarial attacks, especially the role of the early visual system and its influence on the brain manifold. Due to retina gap junctions being crucial for the denoising function in the early visual system, we combine a retina gap junction-based filter, G-filter, with DNN as an abstract human visual system model called the biological hybrid model. We adopt this model to study the defense performance of retina gap junctions and their impact on the brain manifold. Compared with other defense methods, the biological hybrid model is more robust and can be further improved by introducing noise during training. Next, we analyze the manifold and its decision boundary of the biological hybrid model from a geometry perspective. The results show that the biological hybrid model has a unique 2D decision boundary with high nonlinearity and a lower curvature of the decision boundary of the manifold compared to other defense methods. The transforming manifold may account for the high robustness of the biological hybrid model. Finally, to dissect G-filter and clarify its internal mechanism, we borrow the Neural Ordinary Differential Equation (ODE) concept and rewrite G-filter into an equivalent recurrent neural network. The results show that the decision boundary of the model's manifold will gradually change with time and eventually reach a steady state, which is modulated by gap junction conductance, revealing the influence of retina gap junctions on the brain manifold is a gradually evolving process.

Lantian Xue, Yixiong Zou, Peixi Peng et al.

Person Re-identification (Re-ID) has attracted great attention due to its promising real-world applications. However, in practice, it is always costly to annotate the training data to train a Re-ID model, and it still remains challenging to reduce the annotation cost while maintaining the performance for the Re-ID task. To solve this problem, we propose the Annotation Efficient Person Re-Identification method to select image pairs from an alternative pair set according to the fallibility and diversity of pairs, and train the Re-ID model based on the annotation. Specifically, we design an annotation and training framework to firstly reduce the size of the alternative pair set by clustering all images considering the locality of features, secondly select images pairs from intra-/inter-cluster samples for human to annotate, thirdly re-assign clusters according to the annotation, and finally train the model with the re-assigned clusters. During the pair selection, we seek for valuable pairs according to pairs' fallibility and diversity, which includes an intra-cluster criterion to construct image pairs with the most chaotic samples and the representative samples within clusters, an inter-cluster criterion to construct image pairs between clusters based on the second-order Wasserstein distance, and a diversity criterion for clusterbased pair selection. Combining all criteria above, a greedy strategy is developed to solve the pair selection problem. Finally, the above clustering-selecting-annotating-reassigning-training procedure will be repeated until the annotation budget is reached. Extensive experiments on three widely adopted Re-ID datasets show that we can greatly reduce the annotation cost while achieving better performance compared with state-of-the-art works.

Baorui Ma, Haoge Deng, Junsheng Zhou et al.

Text-to-3D generation by distilling pretrained large-scale text-to-image diffusion models has shown great promise but still suffers from inconsistent 3D geometric structures (Janus problems) and severe artifacts. The aforementioned problems mainly stem from 2D diffusion models lacking 3D awareness during the lifting. In this work, we present GeoDream, a novel method that incorporates explicit generalized 3D priors with 2D diffusion priors to enhance the capability of obtaining unambiguous 3D consistent geometric structures without sacrificing diversity or fidelity. Specifically, we first utilize a multi-view diffusion model to generate posed images and then construct cost volume from the predicted image, which serves as native 3D geometric priors, ensuring spatial consistency in 3D space. Subsequently, we further propose to harness 3D geometric priors to unlock the great potential of 3D awareness in 2D diffusion priors via a disentangled design. Notably, disentangling 2D and 3D priors allows us to refine 3D geometric priors further. We justify that the refined 3D geometric priors aid in the 3D-aware capability of 2D diffusion priors, which in turn provides superior guidance for the refinement of 3D geometric priors. Our numerical and visual comparisons demonstrate that GeoDream generates more 3D consistent textured meshes with high-resolution realistic renderings (i.e., 1024 $\times$ 1024) and adheres more closely to semantic coherence.

Jiyuan Zhang, Shiyan Chen, Yajing Zheng et al.

The extraction of a clean background image by removing foreground occlusion holds immense practical significance, but it also presents several challenges. Presently, the majority of de-occlusion research focuses on addressing this issue through the extraction and synthesis of discrete images from calibrated camera arrays. Nonetheless, the restoration quality tends to suffer when faced with dense occlusions or high-speed motions due to limited perspectives and motion blur. To successfully remove dense foreground occlusion, an effective multi-view visual information integration approach is required. Introducing the spike camera as a novel type of neuromorphic sensor offers promising capabilities with its ultra-high temporal resolution and high dynamic range. In this paper, we propose an innovative solution for tackling the de-occlusion problem through continuous multi-view imaging using only one spike camera without any prior knowledge of camera intrinsic parameters and camera poses. By rapidly moving the spike camera, we continually capture the dense stream of spikes from the occluded scene. To process the spikes, we build a novel model \textbf{SpkOccNet}, in which we integrate information of spikes from continuous viewpoints within multi-windows, and propose a novel cross-view mutual attention mechanism for effective fusion and refinement. In addition, we contribute the first real-world spike-based dataset \textbf{S-OCC} for occlusion removal. The experimental results demonstrate that our proposed model efficiently removes dense occlusions in diverse scenes while exhibiting strong generalization.

Shanghang Zhang, Gaole Dai, Tiejun Huang et al.

Rapid advancements in imaging techniques and analytical methods over the past decade have revolutionized our ability to comprehensively probe the biological world at multiple scales, pinpointing the type, quantity, location, and even temporal dynamics of biomolecules. The surge in data complexity and volume presents significant challenges in translating this wealth of information into knowledge. The recently emerged Multimodal Large Language Models (MLLMs) exhibit strong emergent capacities, such as understanding, analyzing, reasoning, and generalization. With these capabilities, MLLMs hold promise to extract intricate information from biological images and data obtained through various modalities, thereby expediting our biological understanding and aiding in the development of novel computational frameworks. Previously, such capabilities were mostly attributed to humans for interpreting and summarizing meaningful conclusions from comprehensive observations and analysis of biological images. However, the current development of MLLMs shows increasing promise in serving as intelligent assistants or agents for augmenting human researchers in biology research

Gaole Dai, Chun-Kai Fan, Yiming Tang et al. · pku

Advances in Parameter-Efficient Fine-Tuning (PEFT) bridged the performance gap with Full Fine-Tuning (FFT) through sophisticated analysis of pre-trained parameter spaces. Starting from drawing insights from Neural Engrams (NE) in Biological Neural Networks (BNNs), we establish a connection between the low-rank property observed during PEFT's parameter space shifting and neurobiological mechanisms. This observation leads to our proposed method, Synapse and Neuron (SAN), which decomposes and propagates scaling components from anterior feature adjusting vectors towards posterior weight matrices. Our approach is theoretically grounded in Long-Term Potentiation/Depression (LTP/D) phenomena, which govern synapse development through neurotransmitter release modulation. Extensive experiments demonstrate its effectiveness: on \textbf{vision tasks} across VTAB, FGVC, and GIC (25 datasets) using ViT, SwinT and ConvNeXt, SAN outperforms FFT up to 8.7% and LoRA by 3.2%; on language tasks using Commonsense Reasoning (8 datasets) with LLaMA models (all generations), surpassing ChatGPT up to 8.5% and LoRA by 4.7%; on visual-language tasks using Mixed Visual Instruction (7 datasets) with LLaVA models, it exceeds FFT up to 2.4% and LoRA by 1.9%. Our code and W&B log will be released.

Muyang He, Yexin Liu, Boya Wu et al.

Multimodal Large Language Models (MLLMs) have demonstrated notable capabilities in general visual understanding and reasoning tasks. However, their deployment is hindered by substantial computational costs in both training and inference, limiting accessibility to the broader research and user communities. A straightforward solution is to leverage smaller pre-trained vision and language models, which inevitably cause significant performance drops. In this paper, we demonstrate the possibility of training a smaller but better MLLM with high-quality training data. Specifically, we introduce Bunny, a family of lightweight MLLMs with flexible vision and language backbones for efficient multimodal learning from selected training data. Experiments show that our Bunny-4B/8B outperforms the state-of-the-art large MLLMs on multiple benchmarks. We expect that this work can provide the community with a clean and flexible open-source tool for further research and development. The code, models, and data can be found in https://github.com/BAAI-DCAI/Bunny.

Fan Bai, Yuxin Du, Tiejun Huang et al.

Medical image analysis is essential to clinical diagnosis and treatment, which is increasingly supported by multi-modal large language models (MLLMs). However, previous research has primarily focused on 2D medical images, leaving 3D images under-explored, despite their richer spatial information. This paper aims to advance 3D medical image analysis with MLLMs. To this end, we present a large-scale 3D multi-modal medical dataset, M3D-Data, comprising 120K image-text pairs and 662K instruction-response pairs specifically tailored for various 3D medical tasks, such as image-text retrieval, report generation, visual question answering, positioning, and segmentation. Additionally, we propose M3D-LaMed, a versatile multi-modal large language model for 3D medical image analysis. Furthermore, we introduce a new 3D multi-modal medical benchmark, M3D-Bench, which facilitates automatic evaluation across eight tasks. Through comprehensive evaluation, our method proves to be a robust model for 3D medical image analysis, outperforming existing solutions. All code, data, and models are publicly available at: https://github.com/BAAI-DCAI/M3D.

Chenyuan Wu, Pengfei Zheng, Ruiran Yan et al.

In this work, we introduce OmniGen2, a versatile and open-source generative model designed to provide a unified solution for diverse generation tasks, including text-to-image, image editing, and in-context generation. Unlike OmniGen v1, OmniGen2 features two distinct decoding pathways for text and image modalities, utilizing unshared parameters and a decoupled image tokenizer. This design enables OmniGen2 to build upon existing multimodal understanding models without the need to re-adapt VAE inputs, thereby preserving the original text generation capabilities. To facilitate the training of OmniGen2, we developed comprehensive data construction pipelines, encompassing image editing and in-context generation data. Additionally, we introduce a reflection mechanism tailored for image generation tasks and curate a dedicated reflection dataset based on OmniGen2. Despite its relatively modest parameter size, OmniGen2 achieves competitive results on multiple task benchmarks, including text-to-image and image editing. To further evaluate in-context generation, also referred to as subject-driven tasks, we introduce a new benchmark named OmniContext. OmniGen2 achieves state-of-the-art performance among open-source models in terms of consistency. We will release our models, training code, datasets, and data construction pipeline to support future research in this field. Project Page: https://vectorspacelab.github.io/OmniGen2; GitHub Link: https://github.com/VectorSpaceLab/OmniGen2

Yijia Guo, Yuanxi Bai, Liwen Hu et al.

We proposed Precomputed RadianceTransfer of GaussianSplats (PRTGS), a real-time high-quality relighting method for Gaussian splats in low-frequency lighting environments that captures soft shadows and interreflections by precomputing 3D Gaussian splats' radiance transfer. Existing studies have demonstrated that 3D Gaussian splatting (3DGS) outperforms neural fields' efficiency for dynamic lighting scenarios. However, the current relighting method based on 3DGS still struggles to compute high-quality shadow and indirect illumination in real time for dynamic light, leading to unrealistic rendering results. We solve this problem by precomputing the expensive transport simulations required for complex transfer functions like shadowing, the resulting transfer functions are represented as dense sets of vectors or matrices for every Gaussian splat. We introduce distinct precomputing methods tailored for training and rendering stages, along with unique ray tracing and indirect lighting precomputation techniques for 3D Gaussian splats to accelerate training speed and compute accurate indirect lighting related to environment light. Experimental analyses demonstrate that our approach achieves state-of-the-art visual quality while maintaining competitive training times and allows high-quality real-time (30+ fps) relighting for dynamic light and relatively complex scenes at 1080p resolution.

Jiyuan Zhang, Kang Chen, Shiyan Chen et al.

Novel View Synthesis plays a crucial role by generating new 2D renderings from multi-view images of 3D scenes. However, capturing high-speed scenes with conventional cameras often leads to motion blur, hindering the effectiveness of 3D reconstruction. To address this challenge, high-frame-rate dense 3D reconstruction emerges as a vital technique, enabling detailed and accurate modeling of real-world objects or scenes in various fields, including Virtual Reality or embodied AI. Spike cameras, a novel type of neuromorphic sensor, continuously record scenes with an ultra-high temporal resolution, showing potential for accurate 3D reconstruction. Despite their promise, existing approaches, such as applying Neural Radiance Fields (NeRF) to spike cameras, encounter challenges due to the time-consuming rendering process. To address this issue, we make the first attempt to introduce the 3D Gaussian Splatting (3DGS) into spike cameras in high-speed capture, providing 3DGS as dense and continuous clues of views, then constructing SpikeGS. Specifically, to train SpikeGS, we establish computational equations between the rendering process of 3DGS and the processes of instantaneous imaging and exposing-like imaging of the continuous spike stream. Besides, we build a very lightweight but effective mapping process from spikes to instant images to support training. Furthermore, we introduced a new spike-based 3D rendering dataset for validation. Extensive experiments have demonstrated our method possesses the high quality of novel view rendering, proving the tremendous potential of spike cameras in modeling 3D scenes.

Pedro R. A. S. Bassi, Wenxuan Li, Yucheng Tang et al.

How can we test AI performance? This question seems trivial, but it isn't. Standard benchmarks often have problems such as in-distribution and small-size test sets, oversimplified metrics, unfair comparisons, and short-term outcome pressure. As a consequence, good performance on standard benchmarks does not guarantee success in real-world scenarios. To address these problems, we present Touchstone, a large-scale collaborative segmentation benchmark of 9 types of abdominal organs. This benchmark is based on 5,195 training CT scans from 76 hospitals around the world and 5,903 testing CT scans from 11 additional hospitals. This diverse test set enhances the statistical significance of benchmark results and rigorously evaluates AI algorithms across various out-of-distribution scenarios. We invited 14 inventors of 19 AI algorithms to train their algorithms, while our team, as a third party, independently evaluated these algorithms on three test sets. In addition, we also evaluated pre-existing AI frameworks--which, differing from algorithms, are more flexible and can support different algorithms--including MONAI from NVIDIA, nnU-Net from DKFZ, and numerous other open-source frameworks. We are committed to expanding this benchmark to encourage more innovation of AI algorithms for the medical domain.

Yiting Dong, Jianhao Ding, Zijie Xu et al.

Spiking Neural Networks (SNNs), with their temporal processing capabilities and biologically plausible dynamics, offer a natural platform for unsupervised representation learning. However, current unsupervised SNNs predominantly employ shallow architectures or localized plasticity rules, limiting their ability to model long-range temporal dependencies and maintain temporal feature consistency. This results in semantically unstable representations, thereby impeding the development of deep unsupervised SNNs for large-scale temporal video data. We propose PredNext, which explicitly models temporal relationships through cross-view future Step Prediction and Clip Prediction. This plug-and-play module seamlessly integrates with diverse self-supervised objectives. We firstly establish standard benchmarks for SNN self-supervised learning on UCF101, HMDB51, and MiniKinetics, which are substantially larger than conventional DVS datasets. PredNext delivers significant performance improvements across different tasks and self-supervised methods. PredNext achieves performance comparable to ImageNet-pretrained supervised weights, through unsupervised training solely on UCF101. Additional experiments demonstrate that PredNext, distinct from forced consistency constraints, substantially improves temporal feature consistency while enhancing network generalization capabilities. This work provides a effective foundation for unsupervised deep SNNs on large-scale temporal video data.

Han Xu, Xuerui Qiu, Baiyu Chen et al.

Current Large Language Models (LLMs) are primarily based on large-scale dense matrix multiplications. Inspired by the brain's information processing mechanism, we explore the fundamental question: how to effectively integrate the brain's spiking-driven characteristics into LLM inference. Spiking Neural Networks (SNNs) possess spike-driven characteristics, and some works have attempted to combine SNNs with Transformers. However, achieving spike-driven LLMs with billions of parameters, relying solely on sparse additions, remains a challenge in the SNN field. To address the issues of limited representational capacity and sparsity in existing spike encoding schemes at the LLM level, we propose SDLLM, a spike-driven large language model that eliminates dense matrix multiplications through sparse addition operations. Specifically, we use the plug-and-play gamma-SQP two-step spike encoding method to ensure that the quantization process aligns with the model's semantic space, mitigating representation degradation caused by binary spikes. Furthermore, we introduce bidirectional encoding under symmetric quantization and membrane potential clipping mechanisms, leading to spike trains with no or low firing counts dominating, significantly reducing the model's spike firing rate, while halving the number of time steps. Experimental results show that SDLLM not only significantly reduces inference costs but also achieves state-of-the-art task performance under the spike-based paradigm. For example, compared to previous spike-based LLMs, SDLLM reduces energy consumption by 7x and improves accuracy by 4.2%. Our model provides inspiration for the architecture design of the next generation of event-driven neuromorphic chips.

Yijia Guo, Wenkai Huang, Tong Hu et al.

3D Gaussian Splatting (3DGS) has recently redefined the paradigm of 3D reconstruction, striking an unprecedented balance between visual fidelity and computational efficiency. As its adoption proliferates, safeguarding the copyright of explicit 3DGS assets has become paramount. However, existing invisible message embedding frameworks struggle to reconcile secure and high-capacity data embedding with intrinsic asset utility, often disrupting the native rendering pipeline or exhibiting vulnerability to structural perturbations. In this work, we present \textbf{\textit{Splats in Splats++}}, a unified and pipeline-agnostic steganography framework that seamlessly embeds high-capacity 3D/4D content directly within the native 3DGS representation. Grounded in a principled analysis of the frequency distribution of Spherical Harmonics (SH), we propose an importance-graded SH coefficient encryption scheme that achieves imperceptible embedding without compromising the original expressive power. To fundamentally resolve the geometric ambiguities that lead to message leakage, we introduce a \textbf{Hash-Grid Guided Opacity Mapping} mechanism. Coupled with a novel \textbf{Gradient-Gated Opacity Consistency Loss}, our formulation enforces a stringent spatial-attribute coupling between the original and hidden scenes, effectively projecting the discrete attribute mapping into a continuous, attack-resilient latent manifold. Extensive experiments demonstrate that our method substantially outperforms existing approaches, achieving up to \textbf{6.28 db} higher message fidelity, \textbf{3$\times$} faster rendering, and exceptional robustness against aggressive 3D-targeted structural attacks (e.g., GSPure). Furthermore, our framework exhibits remarkable versatility, generalizing seamlessly to 2D image embedding, 4D dynamic scene steganography, and diverse downstream tasks.

Shaqi Luo, Yuanyuan Li, Youhao Hu et al.

UMI-style interfaces enable scalable robot learning, but existing systems remain largely visuomotor, relying primarily on RGB observations and trajectory while providing only limited access to physical interaction signals. This becomes a fundamental limitation in contact-rich manipulation, where success depends on contact dynamics such as tactile interaction, internal grasping force, and external interaction wrench that are difficult to infer from vision alone. We present OmniUMI, a unified framework for physically grounded robot learning via human-aligned multimodal interaction. OmniUMI synchronously captures RGB, depth, trajectory, tactile sensing, internal grasping force, and external interaction wrench within a compact handheld system, while maintaining collection--deployment consistency through a shared embodiment design. To support human-aligned demonstration, OmniUMI provides dual-force feedback through bilateral gripper feedback and natural perception of external interaction wrench in the handheld embodiment. Built on this interface, we extend diffusion policy with visual, tactile, and force-related observations, and deploy the learned policy through impedance-based execution for unified regulation of motion and contact behavior. Experiments demonstrate reliable sensing and strong downstream performance on force-sensitive pick-and-place, interactive surface erasing, and tactile-informed selective release. Overall, OmniUMI combines physically grounded multimodal data acquisition with human-aligned interaction, providing a scalable foundation for learning contact-rich manipulation.

Zihan Huang, Xinyu Shi, Zecheng Hao et al. · pku

Spiking neural networks (SNNs) show great potential due to their energy efficiency, fast processing capabilities, and robustness. There are two main approaches to constructing SNNs. Direct training methods require much memory, while conversion methods offer a simpler and more efficient option. However, current conversion methods mainly focus on converting convolutional neural networks (CNNs) to SNNs. Converting Transformers to SNN is challenging because of the presence of non-linear modules. In this paper, we propose an Expectation Compensation Module to preserve the accuracy of the conversion. The core idea is to use information from the previous T time-steps to calculate the expected output at time-step T. We also propose a Multi-Threshold Neuron and the corresponding Parallel Parameter normalization to address the challenge of large time steps needed for high accuracy, aiming to reduce network latency and power consumption. Our experimental results demonstrate that our approach achieves state-of-the-art performance. For example, we achieve a top-1 accuracy of 88.60\% with only a 1\% loss in accuracy using 4 time steps while consuming only 35\% of the original power of the Transformer. To our knowledge, this is the first successful Artificial Neural Network (ANN) to SNN conversion for Spiking Transformers that achieves high accuracy, low latency, and low power consumption on complex datasets. The source codes of the proposed method are available at https://github.com/h-z-h-cell/Transformer-to-SNN-ECMT.

Yijia Guo, Liwen Hu, Yuanxi Bai et al.

3D Gaussian Splatting (3DGS) demonstrates unparalleled superior performance in 3D scene reconstruction. However, 3DGS heavily relies on the sharp images. Fulfilling this requirement can be challenging in real-world scenarios especially when the camera moves fast, which severely limits the application of 3DGS. To address these challenges, we proposed Spike Gausian Splatting (SpikeGS), the first framework that integrates the spike streams into 3DGS pipeline to reconstruct 3D scenes via a fast-moving bio-inspired camera. With accumulation rasterization, interval supervision, and a specially designed pipeline, SpikeGS extracts detailed geometry and texture from high temporal resolution but texture lacking spike stream, reconstructs 3D scenes captured in 1 second. Extensive experiments on multiple synthetic and real-world datasets demonstrate the superiority of SpikeGS compared with existing spike-based and deblur 3D scene reconstruction methods. Codes and data will be released soon.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

Zecheng Hao, Xinyu Shi, Yujia Liu et al. · pku

Compared to traditional Artificial Neural Network (ANN), Spiking Neural Network (SNN) has garnered widespread academic interest for its intrinsic ability to transmit information in a more energy-efficient manner. However, despite previous efforts to optimize the learning algorithm of SNNs through various methods, SNNs still lag behind ANNs in terms of performance. The recently proposed multi-threshold model provides more possibilities for further enhancing the learning capability of SNNs. In this paper, we rigorously analyze the relationship among the multi-threshold model, vanilla spiking model and quantized ANNs from a mathematical perspective, then propose a novel LM-HT model, which is an equidistant multi-threshold model that can dynamically regulate the global input current and membrane potential leakage on the time dimension. The LM-HT model can also be transformed into a vanilla single threshold model through reparameterization, thereby achieving more flexible hardware deployment. In addition, we note that the LM-HT model can seamlessly integrate with ANN-SNN Conversion framework under special initialization. This novel hybrid learning framework can effectively improve the relatively poor performance of converted SNNs under low time latency. Extensive experimental results have demonstrated that our model can outperform previous state-of-the-art works on various types of datasets, which promote SNNs to achieve a brand-new level of performance comparable to quantized ANNs. Code is available at https://github.com/hzc1208/LMHT_SNN.