We present the first unified framework for rate-distortion-optimized compression and segmentation of 3D Gaussian Splatting (3DGS). While 3DGS has proven effective for both real-time rendering and semantic scene understanding, prior works have largely treated these tasks independently, leaving their joint consideration unexplored. Inspired by recent advances in rate-distortion-optimized 3DGS compression, this work integrates semantic learning into the compression pipeline to support decoder-side applications–such as scene editing and manipulation–that extend beyond traditional scene reconstruction and view synthesis. Our scheme features a lightweight implicit neural representation-based hyperprior, enabling efficient entropy coding of both color and semantic attributes while avoiding costly grid-based hyperprior as seen in many prior works. To facilitate compression and segmentation, we further develop compression-guided segmentation learning, consisting of quantization-aware training to enhance feature separability and a quality-aware weighting mechanism to suppress unreliable Gaussian primitives. Extensive experiments on the LERF and 3D-OVS datasets demonstrate that our approach significantly reduces transmission cost while preserving high rendering quality and strong segmentation performance.

Visual instruction tuning (VIT) for large vision-language models (LVLMs) requires training on expansive datasets of image-instruction pairs, which can be costly. Recent efforts in VIT data selection aim to select a small subset of high-quality image-instruction pairs, reducing VIT runtime while maintaining performance comparable to full-scale training. However, a major challenge often overlooked is that generating instructions from unlabeled images1for VIT is highly expensive. Most existing VIT datasets rely heavily on human annotations or paid services like the GPT API, which limits users with constrained resources from creating VIT datasets for custom applications. To address this, we introduce Pre-Instruction Data Selection (PreSel), a more practical data selection paradigm that directly selects the most beneficial unlabeled images and generates instructions only for the selected images. PreSel first estimates the relative importance of each vision task within VIT datasets to derive task-wise sampling budgets. It then clusters image features within each task, selecting the most representative images with the budget. This approach reduces computational overhead for both instruction generation during VIT data formation and LVLM fine-tuning. By generating instructions for only 15% of the images, PreSel achieves performance comparable to full-data VIT on the LLaVA-1.5 and Vision-Flan datasets. The link to our project page: https://bardisafa.github.io/PreSel

Zero-Shot Anomaly Detection (ZSAD) is an emerging AD paradigm. Unlike the traditional unsupervised AD setting that requires a large number of normal samples to train a model, ZSAD is more practical for handling data-restricted real-world scenarios. Recently, Multimodal Large Language Models (MLLMs) have shown revolutionary reasoning capabilities in various vision tasks. However, the reasoning of image abnormalities remains underexplored due to the lack of corresponding datasets and benchmarks. To facilitate research in AD & reasoning, we establish the first visual instruction tuning dataset, Anomaly-Instruct-125k, and the evaluation benchmark, VisA-D&R. Through investigation with our benchmark, we reveal that current MLLMs like GPT-4o cannot accurately detect and describe fine-grained anomalous details in images. To address this, we propose Anomaly-OneVision (Anomaly-OV), the first specialist visual assistant for ZSAD and reasoning. Inspired by human behavior in visual inspection, Anomaly-OV leverages a Look-Twice Feature Matching (LTFM) mechanism to adaptively select and emphasize abnormal visual tokens. Extensive experiments demonstrate that Anomaly-OV achieves significant improvements over advanced generalist models in both detection and reasoning. Extensions to medical and 3D AD are provided for future study. The link to our project page: https://xujiacong.github.io/Anomaly-OV/

This paper presents the first-ever study of adapting compressed image latents to suit the needs of downstream vision tasks that adopt Multimodal Large Language Models (MLLMs). MLLMs have extended the success of large language models to modalities (e.g. images) beyond text, but their billion scale hinders deployment on resource-constrained end devices. While cloud-hosted MLLMs could be available, transmitting raw, uncompressed images captured by end devices to the cloud requires an efficient image compression system. To address this, we focus on emerging neural image compression and propose a novel framework with a lightweight transform-neck and a surrogate loss to adapt compressed image latents for MLLM-based vision tasks. Given the huge scale of MLLMs, our framework excludes the entire downstream MLLM except part of its visual encoder from training our system. This stands out from most existing coding for machine approaches that involve downstream networks in training and thus could be impractical when the networks are MLLMs. The proposed framework is general in that it is applicable to various MLLMs, neural image codecs, and multiple application scenarios, where the neural image codec can be (1) pre-trained for human perception without updating, (2) fully updated for joint human and machine perception, or (3) fully updated for only machine perception. Extensive experiments on different neural image codecs and various MLLMs show that our method achieves great rate-accuracy performance with much less complexity.

Theory-of-mind (ToM) enables humans to infer mental states—such as beliefs, desires, and intentions—forming the foundation of social cognition. Existing computational ToM methods rely on structured workflows with ToM-specific priors or deep model fine-tuning but struggle with scalability in multimodal environments. They remain trapped within the gravitational pull of multi-step planning complexity, failing to generalize as task demands increase. To overcome these limitations, we propose a scalable Bayesian ToM planner. It breaks down ToM complexity into stepwise Bayesian updates. Meanwhile, weak-to-strong control specializes smaller LMs to refine ToM-specific likelihood estimation, transferring their ToM reasoning behavior to larger LMs (7B to 405B) for social and world knowledge integration. This synergistic approach enables scalability, aligning large-model inference with human mental states with Bayesian principles. Extensive experiments demonstrate a 4.6% improvement in accuracy over state-of-the-art methods on multimodal ToM benchmarks, including unseen scenarios, establishing a new standard for modeling human mental states in complex environments.

In the sphere of Cone Beam Computed Tomography (CBCT), acquiring X-ray projections from sufficient angles is indispensable for traditional image reconstruction methods to accurately reconstruct 3D anatomical intricacies. However, this acquisition procedure for the linear accelerator-mounted CBCT systems in radiotherapy takes approximately one minute, impeding its use for ultra-fast intra-fractional motion monitoring during treatment delivery. To address this challenge, we introduce the Patient-specific Cycle-domain Geometric-integrated Denoising Diffusion Probabilistic Model (CG-DDPM). This model aims to leverage patient-specific priors from patient's CT/4DCT images, which are acquired for treatment planning purposes, to reconstruct 3D CBCT from a single-view 2D CBCT projection of any arbitrary angle during treatment, namely single-view reconstructed CBCT (svCBCT). The CG-DDPM framework encompasses a dual DDPM structure: the Projection-DDPM for synthesizing comprehensive full-view projections and the CBCT-DDPM for creating CBCT images. A key innovation is our Cycle-Domain Geometry-Integrated (CDGI) method, incorporating a Cone Beam X-ray Geometric Transformation Module (GTM) to ensure precise, synergistic operation between the dual DDPMs, thereby enhancing reconstruction accuracy and reducing artifacts. Evaluated in a study involving 37 lung cancer patients, the method demonstrated its ability to reconstruct CBCT not only from simulated X-ray projections but also from real-world data. The CG-DDPM significantly outperforms existing V-shape convolutional neural networks (V-nets), Generative Adversarial Networks (GANs), and DDPM methods in terms of reconstruction fidelity and artifact minimization. This was confirmed through extensive voxel-level, structural, visual, and clinical assessments. The capability of CG-DDPM to generate high-quality reconstructed CBCT from a single-view projection at any arbitrary angle using a single model opens the door for ultra-fast, in-treatment volumetric imaging. This is especially beneficial for radiotherapy at motion-associated cancer sites and image-guided interventional procedures.

Deep networks are vulnerable to adversarial examples. Adversarial Training (AT) has been a standard foundation of modern adversarial defense approaches due to its remarkable effectiveness. However, AT is extremely time-consuming, refraining it from wide deployment in practical applications. In this paper, we aim at a non-AT defense: How to design a defense method that gets rid of AT but is still robust against strong adversarial attacks? To answer this question, we resort to adaptive Batch Normalization (BN), inspired by the recent advances in test-time domain adaptation. We propose a novel defense accordingly, referred to as the Adaptive Batch Normalization Network (ABNN). ABNN employs a pre-trained substitute model to generate clean BN statistics and sends them to the target model. The target model is exclusively trained on clean data and learns to align the substitute model’s BN statistics. Experimental results show that ABNN consistently improves adversarial robustness against both digital and physically realizable attacks on both image and video datasets. Furthermore, ABNN can achieve higher clean data performance and significantly lower training time complexity compared to AT-based approaches.

Human action anticipation aims at predicting what people will do in the future based on past observations. In this paper, we introduce Uncertainty-aware Action Decoupling Transformer (UADT) for action anticipation. Unlike existing methods that directly predict action in a verb-noun pair format, we decouple the action anticipation task into verb and noun anticipations separately. The objective is to make the two decoupled tasks assist each other and eventually im-prove the action anticipation task. Specifically, we propose a two-stream Transformer-based architecture which is composed of a verb-to-noun model and a noun-to-verb model. The verb-to-noun model leverages the verb information to improve the noun prediction and the other way around. We extend the model in a probabilistic manner and quantify the predictive uncertainty of each decoupled task to select features. In this way, the noun prediction leverages the most in-formative and redundancy-free verb features and verb pre-diction works similarly. Finally, the two streams are combined dynamically based on their uncertainties to make the joint action anticipation. We demonstrate the efficacy of our method by achieving state-of-the-art performance on action anticipation benchmarks including EPIC-KITCHENS, EGTEA Gaze+, and 50-Salads.

We introduce PlausiVL, a large video-language model for anticipating action sequences that are plausible in the real-world. While significant efforts have been made towards anticipating future actions, prior approaches do not take into account the aspect of plausibility in an action sequence. To address this limitation, we explore the generative capability of a large video-language model in our work and further, develop the understanding of plausibility in an action sequence by introducing two objective functions, a counterfactual-based plausible action sequence learning loss and a long-horizon action repetition loss. We utilize temporal logical constraints as well as verb-noun action pair logical constraints to create implausible/counterfactual action sequences and use them to train the model with plausible action sequence learning loss. This loss helps the model to differentiate between plausible and not plausible action sequences and also helps the model to learn implicit temporal cues crucial for the task of action anticipation. The long-horizon action repetition loss puts a higher penalty on the actions that are more prone to repetition over a longer temporal window. With this penalization, the model is able to generate diverse, plausible action sequences. We evaluate our approach on two large-scale datasets, Ego4D and EPIC-Kitchens-100, and show improvements on the task of action anticipation.

Unsupervised Domain Adaptation (UDA) of semantic segmentation transfers labeled source knowledge to an un-labeled target domain by relying on accessing both the source and target data. However, the access to source data is often restricted or infeasible in real-world scenar-ios. Under the source data restrictive circumstances, UDA is less practical. To address this, recent works have ex-plored solutions under the Source-Free Domain Adaptation (SFDA) setup, which aims to adapt a source-trained model to the target domain without accessing source data. Still, existing SFDA approaches use only image-level information for adaptation, making them sub-optimal in video applications. This paper studies SFDA for Video Semantic Segmentation (VSS), where temporal information is lever-aged to address video adaptation. Specifically, we propose Spatio-Temporal Pixel-Level (STPL) contrastive learning, a novel method that takes full advantage of spatio-temporal information to tackle the absence of source data better. STPL explicitly learns semantic correlations among pixels in the spatio-temporal space, providing strong self-supervision for adaptation to the unlabeled target domain. Extensive experiments show that STPL achieves state-of-the-art performance on VSS benchmarks compared to current UDA and SFDA approaches. Code is available at: https://github.com/shaoyuanlo/STPL

One-class novelty detectors are trained with examples of a particular class and are tasked with identifying whether a query example belongs to the same known class. Most recent advances adopt a deep auto-encoder style architecture to compute novelty scores for detecting novel class data. Deep networks have shown to be vulnerable to adversarial attacks, yet little focus is devoted to studying the adversarial robustness of deep novelty detectors. In this article, we first show that existing novelty detectors are susceptible to adversarial examples. We further demonstrate that commonly-used defense approaches for classification tasks have limited effectiveness in one-class novelty detection. Hence, we need a defense specifically designed for novelty detection. To this end, we propose a defense strategy that manipulates the latent space of novelty detectors to improve the robustness against adversarial examples. The proposed method, referred to as Principal Latent Space (PrincipaLS), learns the incrementally-trained cascade principal components in the latent space to robustify novelty detectors. PrincipaLS can purify latent space against adversarial examples and constrain latent space to exclusively model the known class distribution. We conduct extensive experiments on eight attacks, five datasets and seven novelty detectors, showing that PrincipaLS consistently enhances the adversarial robustness of novelty detection models.

Monocular depth estimation (MDE) has attracted intense study due to its low cost and critical functions for robotic tasks such as localization, mapping and obstacle detection. Supervised approaches have led to great success with the advance of deep learning, but they rely on large quantities of ground-truth depth annotations that are expensive to acquire. Unsupervised domain adaptation (UDA) transfers knowledge from labeled source data to unlabeled target data, so as to relax the constraint of supervised learning. However, existing UDA approaches may not completely align the domain gap across different datasets because of the domain shift problem. We believe better domain alignment can be achieved via well-designed feature decomposition. In this paper, we propose a novel UDA method for MDE, referred to as Learning Feature Decomposition for Adaptation (LFDA), which learns to decompose the feature space into content and style components. LFDA only attempts to align the content component since it has a smaller domain gap. Meanwhile, it excludes the style component which is specific to the source domain from training the primary task. Furthermore, LFDA uses separate feature distribution estimations to further bridge the domain gap. Extensive experiments on three domain adaptative MDE scenarios show that the proposed method achieves superior accuracy and lower computational cost compared to the state-of-the-art approaches.

Adversarial robustness of deep neural networks has been actively investigated. However, most existing defense approaches are limited to a specific type of adversarial perturbations. Specifically, they often fail to offer resistance to multiple attack types simultaneously, i.e., they lack multi-perturbation robustness. Furthermore, compared to image recognition problems, the adversarial robustness of video recognition models is relatively unexplored. While several studies have proposed how to generate adversarial videos, only a handful of approaches about defense strategies have been published in the literature. In this paper, we propose one of the first defense strategies against multiple types of adversarial videos for video recognition. The proposed method, referred to as MultiBN, performs adversarial training on multiple adversarial video types using multiple independent batch normalization (BN) layers with a learning-based BN selection module. With a multiple BN structure, each BN brach is responsible for learning the distribution of a single perturbation type and thus provides more precise distribution estimations. This mechanism benefits dealing with multiple perturbation types. The BN selection module detects the attack type of an input video and sends it to the corresponding BN branch, making MultiBN fully automatic and allowing end-to-end training. Compared to present adversarial training approaches, the proposed MultiBN exhibits stronger multi-perturbation robustness against different and even unforeseen adversarial video types, ranging from Lp-bounded attacks and physically realizable attacks. This holds true on different datasets and target models. Moreover, we conduct an extensive analysis to study the properties of the multiple BN structure.

The majority of adversarial machine learning research focuses on additive attacks, which add adversarial perturbation to input data. On the other hand, unlike image recognition problems, only a handful of attack approaches have been explored in the video domain. In this paper, we propose a novel attack method against video recognition models, Multiplicative Adversarial Videos (MultAV), which imposes perturbation on video data by multiplication. MultAV has different noise distributions to the additive counterparts and thus challenges the defense methods tailored to resisting additive adversarial attacks. Moreover, it can be generalized to not only $\ell_{p}$-norm attacks with a new adversary constraint called ratio bound, but also different types of physically realizable attacks. Experimental results show that the model adversarially trained against additive attack is less robust to MultAV.

Adversarial examples contain carefully crafted perturbations that can fool deep neural networks (DNNs) into making wrong predictions. Enhancing the adversarial robustness of DNNs has gained considerable interest in recent years. Although image transformation-based defenses were widely considered at an earlier time, most of them have been defeated by adaptive attacks. In this paper, we propose a new image transformation defense based on error diffusion halftoning, and combine it with adversarial training to defend against adversarial examples. Error diffusion halftoning projects an image into a 1-bit space and diffuses quantization error to neighboring pixels. This process can remove adversarial perturbations from a given image while maintaining acceptable image quality in the meantime in favor of recognition. Experimental results demonstrate that the proposed method is able to improve adversarial robustness even under advanced adaptive attacks, while most of the other image transformation-based defenses do not. We show that a proper image transformation can still be an effective defense approach. Code: https://github.com/shaoyuanlo/Halftoning-Defense

Adversarial robustness of deep neural networks is an extensively studied problem in the literature and various methods have been proposed to defend against adversarial images. However, only a handful of defense methods have been developed for defending against attacked videos. In this paper, we propose a novel Over-and-Under complete restoration network for Defending against adversarial videos (OUDefend). Most restoration networks adopt an encoder-decoder architecture that first shrinks spatial dimension then expands it back. This approach learns undercomplete representations, which have large receptive fields to collect global information but overlooks local details. On the other hand, overcomplete representations have opposite properties. Hence, OUDefend is designed to balance local and global features by learning those two representations. We attach OUDefend to target video recognition models as a feature restoration block and train the entire network end-to-end. Experimental results show that the defenses focusing on images may be ineffective to videos, while OUDefend enhances robustness against different types of adversarial videos, ranging from additive attacks, multiplicative attacks to physically realizable attacks. Code: https://git;hub.com/shaoyuanlo/OUDefend

Semantic segmentation has made encouraging progress due to the success of deep convolutional networks in recent years. Meanwhile, depth sensors become prevalent nowadays; thus, depth maps can be acquired more easily. However, there are few studies that focus on the RGB-D semantic segmentation task. Exploiting the depth information effectiveness to improve performance is a challenge. In this paper, we propose a novel solution named LDFNet, which incorporates Luminance, Depth and Color information by a fusion-based network. It includes a sub-network to process depth maps and employs luminance images to assist the depth information in processes. LDFNet outperforms the other state-of-art systems on the Cityscapes dataset, and its inference speed is faster than most of the existing networks. The experimental results show the effectiveness of the proposed multi-modal fusion network and its potential for practical applications.

Typical convolutional networks are trained and conducted on RGB images. However, images are often compressed for memory savings and efficient transmission in real-world applications. In this paper, we explore methods for performing semantic segmentation on the discrete cosine transform (DCT) representation defined by the JPEG standard. We first rearrange the DCT coefficients to form a preferred input type, then we tailor an existing network to the DCT inputs. The proposed method has an accuracy close to the RGB model at about the same network complexity. Moreover, we investigate the impact of selecting different DCT components on segmentation performance. With a proper selection, one can achieve the same level accuracy using only 36% of the DCT coefficients. We further show the robustness of our method under the quantization errors. To our knowledge, this paper is the first to explore semantic segmentation on the DCT representation.

Real-time semantic segmentation plays an important role in practical applications such as self-driving and robots. Most semantic segmentation research focuses on improving estimation accuracy with little consideration on efficiency. Several previous studies that emphasize high-speed inference often fail to produce high-accuracy segmentation results. In this paper, we propose a novel convolutional network named Efficient Dense modules with Asymmetric convolution (EDANet), which employs an asymmetric convolution structure and incorporates dilated convolution and dense connectivity to achieve high efficiency at low computational cost and model size. EDANet is 2.7 times faster than the existing fast segmentation network, ICNet, while it achieves a similar mIoU score without any additional context module, post-processing scheme, and pretrained model. We evaluate EDANet on Cityscapes and CamVid datasets, and compare it with the other state-of-art systems. Our network can run with the high-resolution inputs at the speed of 108 FPS on one GTX 1080Ti.

Lane detection plays an important role in a self-driving vehicle. Several studies leverage a semantic segmentation network to extract robust lane features, but few of them can distinguish different types of lanes. In this paper, we focus on the problem of multi-class lane semantic segmentation. Based on the observation that the lane is a small-size and narrow-width object in a road scene image, we propose two techniques, Feature Size Selection (FSS) and Degressive Dilation Block (DD Block). The FSS allows a network to extract thin lane features using appropriate feature sizes. To acquire fine-grained spatial information, the DD Block is made of a series of dilated convolutions with degressive dilation rates. Experimental results show that the proposed techniques provide obvious improvement in accuracy, while they achieve the same or faster inference speed compared to the baseline system, and can run at real-time on high-resolution images.