VideoWorld: Exploring Knowledge Learning from Unlabeled Videos

This work explores whether a deep generative model can learn complex knowledge solely from visual input, in contrast to the prevalent focus on text-based models like large language models (LLMs). We develop VideoWorld, an auto-regressive video generation model trained on unlabeled video data, and test its knowledge acquisition abilities in video-based Go and robotic control tasks. Our experiments reveal two key findings: (1) video-only training provides sufficient information for learning knowledge, including rules, reasoning and planning capabilities, and (2) the representation of visual change is crucial for knowledge acquisition. To improve both the efficiency and efficacy of this process, we introduce the Latent Dynamics Model (LDM) as a key component of VideoWorld. Remarkably, VideoWorld reaches a 5-dan professional level in the Video-GoBench with just a 300-million-parameter model, without relying on search algorithms or reward mechanisms typical in reinforcement learning. In robotic tasks, VideoWorld effectively learns diverse control operations and generalizes across environments, approaching the performance of oracle models in CALVIN and RLBench. This study opens new avenues for knowledge acquisition from visual data, with all code, data, and models open-sourced for further research.

DeepSeek-V3 Technical Report

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.

Revisiting Monocular 3D Object Detection from Scene-Level Depth Retargeting to Instance-Level Spatial Refinement

Monocular 3D object detection is challenging due to the lack of accurate depth. However, existing depth-assisted solutions still exhibit inferior performance, whose reason is universally acknowledged as the unsatisfactory accuracy of monocular depth estimation models. In this paper, we revisit monocular 3D object detection from the depth perspective and formulate an additional issue as the limited 3D structure-aware capability of existing depth representations (\textit{e.g.}, depth one-hot encoding or depth distribution). To address this issue, we propose a novel depth-adapted monocular 3D object detection network, termed \textbf{RD3D}, that mainly comprises a Scene-Level Depth Retargeting (SDR) module and an Instance-Level Spatial Refinement (ISR) module. The former incorporates the scene-level perception of 3D structures, retargeting traditional depth representations to a new formulation: \textbf{Depth Thickness Field}. The latter refines the voxel spatial representation with the guidance of instances, eliminating the ambiguity of 3D occupation and thus improving detection accuracy. Extensive experiments on the KITTI and Waymo datasets demonstrate our superiority to existing state-of-the-art (SoTA) methods and the universality when equipped with different depth estimation models. The code will be available.

CLIP-GS: Unifying Vision-Language Representation with 3D Gaussian Splatting

Recent works in 3D multimodal learning have made remarkable progress. However, typically 3D multimodal models are only capable of handling point clouds. Compared to the emerging 3D representation technique, 3D Gaussian Splatting (3DGS), the spatially sparse point cloud cannot depict the texture information of 3D objects, resulting in inferior reconstruction capabilities. This limitation constrains the potential of point cloud-based 3D multimodal representation learning. In this paper, we present CLIP-GS, a novel multimodal representation learning framework grounded in 3DGS. We introduce the GS Tokenizer to generate serialized gaussian tokens, which are then processed through transformer layers pre-initialized with weights from point cloud models, resulting in the 3DGS embeddings. CLIP-GS leverages contrastive loss between 3DGS and the visual-text embeddings of CLIP, and we introduce an image voting loss to guide the directionality and convergence of gradient optimization. Furthermore, we develop an efficient way to generate triplets of 3DGS, images, and text, facilitating CLIP-GS in learning unified multimodal representations. Leveraging the well-aligned multimodal representations, CLIP-GS demonstrates versatility and outperforms point cloud-based models on various 3D tasks, including multimodal retrieval, zero-shot, and few-shot classification.

Unsupervised Region-Based Image Editing of Denoising Diffusion Models

Although diffusion models have achieved remarkable success in the field of image generation, their latent space remains under-explored. Current methods for identifying semantics within latent space often rely on external supervision, such as textual information and segmentation masks. In this paper, we propose a method to identify semantic attributes in the latent space of pre-trained diffusion models without any further training. By projecting the Jacobian of the targeted semantic region into a low-dimensional subspace which is orthogonal to the non-masked regions, our approach facilitates precise semantic discovery and control over local masked areas, eliminating the need for annotations. We conducted extensive experiments across multiple datasets and various architectures of diffusion models, achieving state-of-the-art performance. In particular, for some specific face attributes, the performance of our proposed method even surpasses that of supervised approaches, demonstrating its superior ability in editing local image properties.

Memory Efficient Matting with Adaptive Token Routing

Transformer-based models have recently achieved outstanding performance in image matting. However, their application to high-resolution images remains challenging due to the quadratic complexity of global self-attention. To address this issue, we propose MEMatte, a \textbf{m}emory-\textbf{e}fficient \textbf{m}atting framework for processing high-resolution images. MEMatte incorporates a router before each global attention block, directing informative tokens to the global attention while routing other tokens to a Lightweight Token Refinement Module (LTRM). Specifically, the router employs a local-global strategy to predict the routing probability of each token, and the LTRM utilizes efficient modules to simulate global attention. Additionally, we introduce a Batch-constrained Adaptive Token Routing (BATR) mechanism, which allows each router to dynamically route tokens based on image content and the stages of attention block in the network. Furthermore, we construct an ultra high-resolution image matting dataset, UHR-395, comprising 35,500 training images and 1,000 test images, with an average resolution of $4872\times6017$. This dataset is created by compositing 395 different alpha mattes across 11 categories onto various backgrounds, all with high-quality manual annotation. Extensive experiments demonstrate that MEMatte outperforms existing methods on both high-resolution and real-world datasets, significantly reducing memory usage by approximately 88% and latency by 50% on the Composition-1K benchmark. Our code is available at https://github.com/linyiheng123/MEMatte.

CSR:Achieving 1 Bit Key-Value Cache via Sparse Representation

The emergence of long-context text applications utilizing large language models (LLMs) has presented significant scalability challenges, particularly in memory footprint. The linear growth of the Key-Value (KV) cache responsible for storing attention keys and values to minimize redundant computations can lead to substantial increases in memory consumption, potentially causing models to fail to serve with limited memory resources. To address this issue, we propose a novel approach called Cache Sparse Representation (CSR), which converts the KV cache by transforming the dense Key-Value cache tensor into sparse indexes and weights, offering a more memory-efficient representation during LLM inference. Furthermore, we introduce NeuralDict, a novel neural network-based method for automatically generating the dictionary used in our sparse representation. Our extensive experiments demonstrate that CSR achieves performance comparable to state-of-the-art KV cache quantization algorithms while maintaining robust functionality in memory-constrained environments.

Behavior Backdoor for Deep Learning Models

The various post-processing methods for deep-learning-based models, such as quantification, pruning, and fine-tuning, play an increasingly important role in artificial intelligence technology, with pre-train large models as one of the main development directions. However, this popular series of post-processing behaviors targeting pre-training deep models has become a breeding ground for new adversarial security issues. In this study, we take the first step towards ``behavioral backdoor'' attack, which is defined as a behavior-triggered backdoor model training procedure, to reveal a new paradigm of backdoor attacks. In practice, we propose the first pipeline of implementing behavior backdoor, i.e., the Quantification Backdoor (QB) attack, upon exploiting model quantification method as the set trigger. Specifically, to adapt the optimization goal of behavior backdoor, we introduce the behavior-driven backdoor object optimizing method by a bi-target behavior backdoor training loss, thus we could guide the poisoned model optimization direction. To update the parameters across multiple models, we adopt the address-shared backdoor model training, thereby the gradient information could be utilized for multimodel collaborative optimization. Extensive experiments have been conducted on different models, datasets, and tasks, demonstrating the effectiveness of this novel backdoor attack and its potential application threats.

BGM: Background Mixup for X-ray Prohibited Items Detection

Prohibited item detection is crucial for ensuring public safety, yet current X-ray image-based detection methods often lack comprehensive data-driven exploration. This paper introduces a novel data augmentation approach tailored for prohibited item detection, leveraging unique characteristics inherent to X-ray imagery. Our method is motivated by observations of physical properties including: 1) X-ray Transmission Imagery: Unlike reflected light images, transmitted X-ray pixels represent composite information from multiple materials along the imaging path. 2) Material-based Pseudo-coloring: Pseudo-color rendering in X-ray images correlates directly with material properties, aiding in material distinction. Building on a novel perspective from physical properties, we propose a simple yet effective X-ray image augmentation technique, Background Mixup (BGM), for prohibited item detection in security screening contexts. The essence is the rich background simulation of X-ray images to induce the model to increase its attention to the foreground. The approach introduces 1) contour information of baggage and 2) variation of material information into the original image by Mixup at patch level. Background Mixup is plug-and-play, parameter-free, highly generalizable and provides an effective solution to the limitations of classical visual augmentations in non-reflected light imagery. When implemented with different high-performance detectors, our augmentation method consistently boosts performance across diverse X-ray datasets from various devices and environments. Extensive experimental results demonstrate that our approach surpasses strong baselines while maintaining similar training resources.

ROSE: Revolutionizing Open-Set Dense Segmentation with Patch-Wise Perceptual Large Multimodal Model

Advances in CLIP and large multimodal models (LMMs) have enabled open-vocabulary and free-text segmentation, yet existing models still require predefined category prompts, limiting free-form category self-generation. Most segmentation LMMs also remain confined to sparse predictions, restricting their applicability in open-set environments. In contrast, we propose ROSE, a Revolutionary Open-set dense SEgmentation LMM, which enables dense mask prediction and open-category generation through patch-wise perception. Our method treats each image patch as an independent region of interest candidate, enabling the model to predict both dense and sparse masks simultaneously. Additionally, a newly designed instruction-response paradigm takes full advantage of the generation and generalization capabilities of LMMs, achieving category prediction independent of closed-set constraints or predefined categories. To further enhance mask detail and category precision, we introduce a conversation-based refinement paradigm, integrating the prediction result from previous step with textual prompt for revision. Extensive experiments demonstrate that ROSE achieves competitive performance across various segmentation tasks in a unified framework. Code will be released.