We have witnessed promising progress led by large language models (LLMs) and further vision language models (VLMs) in handling various queries as a general-purpose assistant. VLMs, as a bridge to connect the visual world and language corpus, receive both visual content and various text-only user instructions to generate corresponding responses. Though great success has been achieved by VLMs in various fields, in this work, we ask whether the existing VLMs can act as domain experts, accurately answering marine questions, which require significant domain expertise and address special domain challenges/requirements. To comprehensively evaluate the effectiveness and explore the boundary of existing VLMs, we construct the first large-scale marine VLM dataset and benchmark called MarineEval, with 2,000 image-based question-answering pairs. During our dataset construction, we ensure the diversity and coverage of the constructed data: 7 task dimensions and 20 capacity dimensions. The domain requirements are specially integrated into the data construction and further verified by the corresponding marine domain experts. We comprehensively benchmark 17 existing VLMs on our MarineEval and also investigate the limitations of existing models in answering marine research questions. The experimental results reveal that existing VLMs cannot effectively answer the domain-specific questions, and there is still a large room for further performance improvements. We hope our new benchmark and observations will facilitate future research. Project Page: http://marineeval.hkustvgd.com/

Multimodal recommendation systems (MRS) jointly model user-item interaction graphs and rich item content, but this tight coupling makes user data difficult to remove once learned. Approximate machine unlearning offers an efficient alternative to full retraining, yet current MRS unlearning applies reverse updates largely uniformly across model components. We show that this uniform treatment is misaligned with modern MRS: deleted-data influence is distributed unevenly across ranking behavior, modality branches, and model modules. This non-uniformity gives rise to three bottlenecks in MRS unlearning: target-item persistence in the collaborative graph, modality imbalance across feature branches, and concentrated module-level sensitivity in the parameter space. To address this mismatch, we propose targeted reverse update (TRU), a plug-and-play unlearning framework for MRS. Instead of applying a uniform global reversal, TRU performs three coordinated interventions across the model hierarchy: a ranking fusion gate to suppress residual target-item influence in ranking, branch-wise modality scaling to preserve retained multimodal representations, and capacity-aware parameter-group selection to localize reverse updates to deletion-sensitive modules. Across two backbones, three datasets, and three unlearning regimes, TRU achieves a stronger retain–forget trade-off than MMRecUn in most settings. In two challenging user-level cases, TRU also attains favorable operating points among all evaluated baselines. Security audits report the lowest MIA BalAcc and a tie for the lowest ASR among approximate methods in both audited settings, while wall-clock trajectories show earlier convergence to favorable retain–forget regions.

The rapid advancement of generative technologies has made synthetic images nearly indistinguishable from real ones, thereby creating an urgent need for robust detectors to counter misinformation. However, existing methods mainly rely on universal artifact features that are shared across multiple generators. We observe that as the diversity of generators increases, the overlap of these common features gradually decreases. This severely undermines model generalization. In contrast, focusing only on unique artifacts tends to cause overfitting to specific forgery patterns. To address this challenge, we propose LEGO (LoRA-Enabled Generator-Oriented Framework). The core mechanism of LEGO employs an MLP to modulate multiple LoRA (Low-Rank Adaptation) blocks, each pretrained to capture the unique artifacts of a specific generator, followed by attention-based feature fusion. Unlike conventional methods that seek a single universal solution, LEGO delegates unique artifact extraction to specialized LoRA modules by dividing its training procedure into two stages. Each LoRA module is individually trained on a single-generator dataset to learn generator-specific representations, then MLP and attention layers are trained on mixed datasets to dynamically regulate the contribution of each module. Benefiting from its modular yet robust design, LEGO can be naturally extended by incorporating new LoRA modules for adaptation to newly emerging next-generation datasets, while still achieving substantially better performance than prior SOTA methods with fewer than 30,000 training images, less than 10

Automatically identifying plankton categories using deep learning (DL) and computer vision (CV) is of great significance for biodiversity research and ecological environmental protection. The current automatic identification system is mainly used as supporting software for instruments and equipment. However, they can only identify a limited number of species in datasets collected in specific environments, so their application focus is single and flexible, making it difficult to meet the needs of emerging technologies in the context of rapid development of artificial intelligence. This paper designs and implements the “PlanktonTaxonPred” foundation model system, a Plankton Taxonomy Prediction system based on the vision-language model (VLM), which covers the complete process of data acquisition, data preprocessing, algorithm model embedding, and application system, and realizes the fine prediction of hierarchical taxonomic information of more than 200 types of plankton (including phytoplankton and zooplankton). The PlanktonTaxonPred system can provide an effective algorithmic model and application support for related scientific research and the intelligent upgrading of instruments and equipment.

Underwater robotic systems increasingly rely on high-resolution video, but practical enhancement must jointly satisfy restoration quality, latency, and resource constraints. Frame-wise methods cannot exploit inter-frame redundancy, whereas recurrent and sliding-window schemes introduce sequential or repeated computation, making real-time 4K processing difficult. We propose the Efficient Underwater Video Enhancement Model (EUVM) for high-quality and efficient underwater video enhancement. Its core is Pre-Reshape, an architecture-level computation organization strategy that transforms an input clip into a computation-friendly, channel-expanded, and spatially compact representation. This organization enables all frames in the clip to be processed jointly in one pass on compact intermediate feature maps. A redundancy-aware compression layer further compacts repeated multi-frame information and reduces the cost of the subsequent enhancement backbone. To preserve degradation-aware restoration under this efficient representation, the Distance-Related Compensation Module and Attenuation-Related Compensation Module learn lightweight feature-space factors associated with distance-related spatial variation and wavelength-dependent channel attenuation, without explicitly recovering physical variables. A Reshape-based State-Space Module further adapts long-range modeling to the reshaped layout by restoring local interactions before global state-space propagation. Experiments on UVEB, DRUVA, UIEB, and LSUI demonstrate competitive enhancement quality across diverse underwater conditions. On UVEB, EUVM achieves 26.31 dB PSNR while processing 4K inputs at 25.5 FPS on an NVIDIA A40 GPU. With graph partitioning and mixed-precision embedded execution, EUVM further reaches 20.83 output FPS at 2K on Jetson Orin NX while preserving enhancement quality close to the server-side reference.

Reliable feedforward underwater 3D reconstruction remains challenging due to severe light attenuation and backscattering, which degrade visual quality and disrupt feature consistency across views, leading to inaccurate multi-view geometry. To address this issue, we propose WAT3R, a feed-forward framework for reconstructing 3D scenes directly from underwater images. By leveraging degradation adaptation as a geometry-constrained process, WAT3R integrates a lightweight neural adaptation module to flexibly account for these underwater imaging effects, thereby improving multi-view reconstruction quality. Implemented in a single forward pass, WAT3R directly and efficiently outputs pixel-aligned 3D point maps and camera poses from underwater videos, allowing a high-quality underwater 3D reconstruction. Experiments conducted on the FLSea, SQUID, and USOD10K datasets show that our method consistently outperforms state-of-the-art approaches on 3D reconstruction tasks, including multi-view/monocular depth estimation and camera pose estimation.

We have been witnessing remarkable success led by the power of neural networks driven by a significant scale of training data in handling various computer vision tasks. However, less attention has been paid to monitoring the camouflaged animals, the masters of hiding themselves in the background. Robust and precise segmentation of camouflaged animals is challenging even for domain experts due to their similarity to the environment. Although several efforts have been made in camouflaged animal image segmentation, to the best of our knowledge, limited work exists on camouflaged animal video understanding (CAVU). Biologists often prefer videos for monitoring and understanding animal behaviors, as videos provide redundant information and temporal consistency. However, the scarcity of labeled video data significantly hinders progress in this area. To address these challenges, we present CamoVid60K, a diverse, large-scale, and accurately annotated video dataset of camouflaged animals. This dataset comprises 218 videos with 62,774 finely annotated frames, covering 70 animal categories, which surpasses all previous datasets in terms of the number of videos/frames and species included. CamoVid60K also offers more diverse downstream tasks in computer vision, such as camouflaged animal classification, detection, and task-specific segmentation (semantic, referring, motion),etc.We have benchmarked several state-of-the-art algorithms on the proposed CamoVid60K dataset, and the experimental results provide valuable insights for future research directions. Our dataset serves as a novel and challenging benchmark to stimulate the development of more powerful camouflaged animal video segmentation algorithms, with substantial room for further improvement.

Marine visual understanding is essential for monitoring and protecting marine ecosystems, enabling automatic and scalable biological surveys. However, progress is hindered by limited training data and the lack of a systematic task formulation that aligns domain-specific marine challenges with well-defined computer vision tasks, thereby limiting effective model application. To address this gap, we present ORCA, a multi-modal benchmark for marine research comprising 14,647 images from 478 species, with 42,217 bounding box annotations and 22,321 expert-verified instance captions. The dataset provides fine-grained visual and textual annotations that capture morphology-oriented attributes across diverse marine species. To catalyze methodological advances, we evaluate 18 state-of-the-art models on three tasks: object detection (closed-set and open-vocabulary), instance captioning, and visual grounding. Results highlight key challenges, including species diversity, morphological overlap, and specialized domain demands, underscoring the difficulty of marine understanding. ORCA thus establishes a comprehensive benchmark to advance research in marine domain. Project Page: http://orca.hkustvgd.com/.

Glass is a prevalent material among solid objects in everyday life, yet segmentation methods struggle to distinguish it from opaque materials due to its transparency and reflection. While it is known that human perception relies on boundary and reflective-object features to distinguish glass objects, the existing literature has not yet sufficiently captured both properties when handling transparent objects. Hence, we propose incorporating both of these powerful visual cues via the Boundary Feature Enhancement and Reflection Feature Enhancement modules in a mutually beneficial way. Our proposed framework, TransCues, is a pyramidal transformer encoder-decoder architecture to segment transparent objects. We empirically show that these two modules can be used together effectively, improving overall performance across various benchmark datasets, including glass object semantic segmentation, mirror object semantic segmentation, and generic segmentation datasets. Our method outperforms the state-of-the-art by a large margin, achieving +4.2

The growing demand for efficient image segmentation in marine ecological studies is currently constrained by two key factors: the high computational requirements of models such as the Segment Anything Model (SAM) and the degraded accuracy of lightweight models in underwater environments. To overcome these challenges, we introduce a Three-stage Iterative Optimization (TIO) training framework and a decoupled knowledge distillation strategy, termed MaskGuide, both of which facilitate the development of our optimized Tiny-MSAM model on the edge device. This model significantly improves frame processing speed while maintaining high accuracy, achieving an optimal tradeoff between segmentation precision and inference speed in practical underwater and marine applications. Our research provides a feasible direction for deploying advanced segmentation models in marine and other resource-constrained scenarios. The Tiny-MSAM model, even when trained from scratch, contains only 0.5% of the parameters of the SAM model and 58% of those in MobileSAM. In existing underwater image segmentation benchmarks (e.g., UIIS dataset), it outperforms MobileSAM by a large margin and reaches 99.4% of the performance of the SAM ViT-H variant.

Domain Adaptive Object Detection (DAOD) addresses the challenge of transferring object detectors from labeled source domains to unlabeled target domains. Existing domain adaptation methods primarily rely on feature distribution alignment, which enhances domain-invariant features (statistical invariance) but also inadvertently increases inherent domain-common spurious factors (e.g., common environmental contexts), which act as shortcut features rather than the true causal factors for object classification. We propose Dynamic Causal Refinement (DCR), a novel framework that establishes a closed-loop feedback mechanism between data augmentation and model optimization to progressively refine causal features. Specifically, we design Semantic Prediction Consistency (SPC) to filter domain-specific spurious factors and establish a robust statistical invariance, and Discrepancy-Guided Causal Refinement (DGCR) to actively suppress the dependence on domain-common spurious factors via spectral perturbation for causal refinement. This process encourages the detector to suppress its reliance on shortcut features and instead prioritize semantically meaningful causal representations. Extensive experiments on standard benchmarks demonstrate that our method outperforms state-of-the-art counterparts significantly.

Recent Vision-Language Models (VLMs) have achieved remarkable success in visual understanding, driven by the growing availability of high-quality image-text pairs. However, the performance of VLMs often degrades in the video domain due to the essential need for temporal understanding and the scarcity of large-scale annotated video data. In this work, we focus on marine video understanding, which brings further challenges: first, it requires substantial domain expertise; and video VLMs usually struggle with localizing and interpreting critical information from marine videos, as the informative events are typically sparse, unpredictable, and unevenly distributed. To address these challenges, we carefully curate the first event-centric marine video understanding dataset called MarineEVT, which features 20K multi-task, video-level visual question-answering pairs spanning multiple dimensions of marine understanding and analysis. Meanwhile, based on MarineEVT, we decompose marine video understanding as an Event-centric Visual Tool-integrated Reasoning process EVT-R1 for short, where we leverage powerful visual tools to drive the model to localize and interpret critical information aligned with visual questions and human intent. To demonstrate its effectiveness, we compare EVT-R1 against 11 SOTA VLMs in different settings. EVT-R1 outperforms the top open-source and top commercial models by 5.22 and 11.09, respectively. MarineEVT and EVT-R1 lay the foundation for ecological discovery and marine education, fostering the development of VLMs capable of interpreting marine dynamics, reasoning about ecological interactions, and supporting sustainable ocean video understanding and analysis.

Federated graph learning (FGL) trains a shared graph model across clients whose local graphs differ in node features, labels, and connectivity while keeping raw graph data decentralized. Although graph-domain shifts across clients can severely degrade the global model, existing FGL approaches for graph-domain shift mainly adapt local representations, propagation, or graph-derived collaboration, while the server typically applies the same aggregation rule to every parameter coordinate. In this work, we analyze how graph-domain shifts affect server-side aggregation. We find that, as clients optimize under distinct graph-domain conditions, they gradually concentrate their strongest updates on different parameter coordinates, making important update coordinates less shared across clients. Consequently, standard averaging can weaken these domain-specific signals even when client updates are not directly opposed. These findings reveal an aggregationinduced signal dilution effect: parameter coordinates strongly expressed by only a subset of domains are attenuated by nearzero contributions from the remaining domains. Motivated by these findings, we propose FedIA, a lightweight serverside aggregation method that calibrates parameter selection and client weighting. FedIA identifies important coordinates =within each layer and adjusts client aggregation weights according to their contributions on the selected subspace, without modifying local graph training or the standard clientupdate payload. Experiments on Twitch Gamers and WikiNet using two graph-learning backbones and nine FL/FGL baselines show improvements of up to 11.58 percentage points, with particularly strong gains under domain skew.

Coral reef imagery offers critical data for monitoring ecosystem health, in particular as the ease of image datasets continues to rapidly expand. Whilst semi-automated analytical platforms for reef imagery are becoming more available, the dominant approaches face fundamental limitations. To address these challenges, we propose CoralSCOP-LAT, a coral reef image analysis and labeling tool that automatically segments and analyzes coral regions. By leveraging advanced machine learning models tailored for coral reef segmentation, CoralSCOPLAT enables users to generate dense segmentation masks with minimal manual effort, significantly enhancing both the labeling efficiency and precision of coral reef analysis. Our extensive evaluations demonstrate that CoralSCOP-LAT surpasses existing coral reef analysis tools in terms of time efficiency, accuracy, precision, and flexibility. CoralSCOP-LAT, therefore, not only accelerates the coral reef annotation process but also assists users in obtaining high-quality coral reef segmentation and analysis outcomes.

Visual images corrupted by various types and levels of degradations are commonly encountered in practical image compression. However, most existing image compression methods are tailored for clean images, therefore struggling to achieve satisfying results on these images. Joint compression and restoration methods typically focus on a single type of degradation and fail to address a variety of degradations in practice. To this end, we propose a unified framework for all-in-one image compression and restoration, which incorporates the image restoration capability against various degradations into the process of image compression. The key challenges involve distinguishing authentic image content from degradations, and flexibly eliminating various degradations without prior knowledge. Specifically, the proposed framework approaches these challenges from two perspectives: i.e., content information aggregation, and degradation representation aggregation. Extensive experiments demonstrate the following merits of our model: 1) superior rate-distortion (RD) performance on various degraded inputs while preserving the performance on clean data; 2) strong generalization ability to real-world and unseen scenarios; 3) higher computing efficiency over compared methods. Our code is available at https://github.com/ZeldaM1/All-in-one.

Underwater coral reef monitoring plays an important role in the maintenance and protection of the underwater ecosystem. Extracting information from the collected coral reef images and videos based on computer vision techniques has recently gained increasing attention. Semantic segmentation, which assigns semantic category information to each pixel in images, has been introduced to understand coral reefs. Satisfactory semantic segmentation performance has been achieved based on large-scale in-air data sets with densely labeled annotations. However, underwater coral reef understanding is less explored and existing underwater coral reef data sets are mainly captured under ideal and normal conditions and lack variance. They cannot fully reflect the diversity and properties of coral reefs. Thus, trained coral reef segmentation models show very limited performance when deployed in practical, challenging, and adverse conditions. To address these issues, in this article, we propose an in-the-wild coral reef data set named HKCoral to close the gap for performing in-situ coral reef monitoring. The collected data set with dense pixel-wise annotations possesses larger diversity, appearance, viewpoint, and visibility variations. Besides, we adopt the fundamental coral growth form as the foundation of our semantic coral reef segmentation, which enables a strong generalizability to unseen coral reef images from different sites. We benchmark the coral reef segmentation performance of 17 state-of-the-art semantic segmentation algorithms (including the recent generalist segment anything model) and further introduce a complementary architecture to better utilize underwater image enhancement for improving the segmentation performance of models. We have conducted extensive experiments based on various up-to-date segmentation models on our benchmark and the experimental results demonstrate that there is still ample room to improve coral segmentation performance. Ablation studies and discussions are also included. The proposed benchmark could significantly enhance the efficiency and accuracy of real-world underwater coral reef surveying.

We investigate coral reef semantic segmentation, in which multifaceted factors, like genes, environmental changes, and internal interactions, can lead to highly unpredictable growth patterns. Existing segmentation approaches in both computer vision and coral reef communities have failed to incorporate the intrinsic properties of corals, specifically their self-repeated, asymmetric, and amorphous distribution of elements, into model design. We propose CoralSRT, a feature rectification module via self-supervised guidance, to reduce the stochasticity of coral features extracted by pretrained foundation models (FMs), as demonstrated in Fig. 1. Our insight is that while different corals are highly dissimilar, individual corals within the same growth exhibit strong selfaffinity. Using a superset of features from FMs learned by various pretext tasks, we extract a pattern related to the intrinsic properties of each coral to strengthen within-segment affinity, aligning with centrality. We investigate features from FMs that were optimized by various pretext tasks on significantly large-scale unlabeled or labeled data, which already contain rich information for modeling both within-segment and cross-segment affinities, enabling the adaptation of FMs for coral segmentation. CoralSRT can rectify features from FMs to more efficient features for label propagation and lead to further significant semantic segmentation perfor-mance gains, all without requiring additional human super-vision, retraining/finetuning FMs or even domain-specific data. These advantages help reduce human effort and the need for domain expertise in data collection and labeling. Our method is easy to implement, and also task- and model-agnostic. CoralSRT bridges the self-supervised pre-training and supervised training in the feature space, also offering insights for segmenting elements/stuffs (e.g., grass, plants, cells, and biofoulings).

Although glass is a prevalent material in everyday life, most semantic segmentation methods struggle to distinguish it from opaque materials. We propose $\textbf{TransCues}$, a pyramidal transformer encoder-decoder architecture to segment transparent objects from a color image. To distinguish between glass and non-glass regions, our transformer architecture is based on two important visual cues that involve boundary and reflection feature learning, respectively. We implement this idea by introducing a Boundary Feature Enhancement (BFE) module paired with a boundary loss and a Reflection Feature Enhancement (RFE) module that decomposes reflections into foreground and background layers. We empirically show that these two modules can be used together effectively, leading to improved overall performance on various benchmark datasets. In addition to binary segmentation of glass and mirror objects, we further demonstrate that our method works well for generic semantic segmentation for both glass and non-glass labels. Our method outperforms the state-of-the-art methods by a large margin on diverse datasets, achieving $\textbf{+4.2}$\% mIoU on Trans10K-v2, $\textbf{+5.6}$\% mIoU on MSD, $\textbf{+10.1}$\% mIoU on RGBD-Mirror, $\textbf{+13.1}$\% mIoU on TROSD, and $\textbf{+8.3}$\% mIoU on Stanford2D3D, demonstrate the effectiveness and efficiency of our method.

Large Language Models (LLMs) demonstrate strong proficiency in generating code for high-resource programming languages (HRPLs) like Python but struggle significantly with low-resource programming languages (LRPLs) such as Racket or D. This performance gap deepens the digital divide, preventing developers using LRPLs from benefiting equally from LLM advancements and reinforcing disparities in innovation within underrepresented programming communities. While generating additional training data for LRPLs is promising, it faces two key challenges: manual annotation is labor-intensive and costly, and LLM-generated LRPL code is often of subpar quality. The underlying cause of this issue is the gap between natural language to programming language gap (NL-PL Gap), which is especially pronounced in LRPLs due to limited aligned data. In this work, we introduce a novel approach called Bridge-Coder, which leverages LLMs' intrinsic capabilities to enhance the performance on LRPLs. Our method consists of two key stages. Bridge Generation, where we create high-quality dataset by utilizing LLMs' general knowledge understanding, proficiency in HRPLs, and in-context learning abilities. Then, we apply the Bridged Alignment, which progressively improves the alignment between NL instructions and LRPLs. Experimental results across multiple LRPLs show that Bridge-Coder significantly enhances model performance, demonstrating the effectiveness and generalization of our approach. Furthermore, we offer a detailed analysis of the key components of our method, providing valuable insights for future work aimed at addressing the challenges associated with LRPLs.