Taming SAM3 in the Wild: A Concept Bank for Open-Vocabulary Segmentation

The recent introduction of \texttt{SAM3} has revolutionized Open-Vocabulary Segmentation (OVS) through \textit{promptable concept segmentation}, which grounds pixel predictions in flexible concept prompts. However, this reliance on pre-defined concepts makes the model vulnerable: when visual distributions shift (\textit{data drift}) or conditional label distributions evolve (\textit{concept drift}) in the target domain, the alignment between visual evidence and prompts breaks down. In this work, we present \textsc{ConceptBank}, a parameter-free calibration framework to restore this alignment on the fly. Instead of adhering to static prompts, we construct a dataset-specific concept bank from the target statistics. Our approach (\textit{i}) anchors target-domain evidence via class-wise visual prototypes, (\textit{ii}) mines representative supports to suppress outliers under data drift, and (\textit{iii}) fuses candidate concepts to rectify concept drift. We demonstrate that \textsc{ConceptBank} effectively adapts \texttt{SAM3} to distribution drifts, including challenging natural-scene and remote-sensing scenarios, establishing a new baseline for robustness and efficiency in OVS. Code and model are available at https://github.com/pgsmall/ConceptBank.

Combating Noisy Labels through Fostering Self- and Neighbor-Consistency

Label noise is pervasive in various real-world scenarios, posing challenges in supervised deep learning. Deep networks are vulnerable to such label-corrupted samples due to the memorization effect. One major stream of previous methods concentrates on identifying clean data for training. However, these methods often neglect imbalances in label noise across different mini-batches and devote insufficient attention to out-of-distribution noisy data. To this end, we propose a noise-robust method named Jo-SNC (\textbf{Jo}int sample selection and model regularization based on \textbf{S}elf- and \textbf{N}eighbor-\textbf{C}onsistency). Specifically, we propose to employ the Jensen-Shannon divergence to measure the ``likelihood'' of a sample being clean or out-of-distribution. This process factors in the nearest neighbors of each sample to reinforce the reliability of clean sample identification. We design a self-adaptive, data-driven thresholding scheme to adjust per-class selection thresholds. While clean samples undergo conventional training, detected in-distribution and out-of-distribution noisy samples are trained following partial label learning and negative learning, respectively. Finally, we advance the model performance further by proposing a triplet consistency regularization that promotes self-prediction consistency, neighbor-prediction consistency, and feature consistency. Extensive experiments on various benchmark datasets and comprehensive ablation studies demonstrate the effectiveness and superiority of our approach over existing state-of-the-art methods.

AbductiveMLLM: Boosting Visual Abductive Reasoning Within MLLMs

Visual abductive reasoning (VAR) is a challenging task that requires AI systems to infer the most likely explanation for incomplete visual observations. While recent MLLMs develop strong general-purpose multimodal reasoning capabilities, they fall short in abductive inference, as compared to human beings. To bridge this gap, we draw inspiration from the interplay between verbal and pictorial abduction in human cognition, and propose to strengthen abduction of MLLMs by mimicking such dual-mode behavior. Concretely, we introduce AbductiveMLLM comprising of two synergistic components: REASONER and IMAGINER. The REASONER operates in the verbal domain. It first explores a broad space of possible explanations using a blind LLM and then prunes visually incongruent hypotheses based on cross-modal causal alignment. The remaining hypotheses are introduced into the MLLM as targeted priors, steering its reasoning toward causally coherent explanations. The IMAGINER, on the other hand, further guides MLLMs by emulating human-like pictorial thinking. It conditions a text-to-image diffusion model on both the input video and the REASONER's output embeddings to "imagine" plausible visual scenes that correspond to verbal explanation, thereby enriching MLLMs' contextual grounding. The two components are trained jointly in an end-to-end manner. Experiments on standard VAR benchmarks show that AbductiveMLLM achieves state-of-the-art performance, consistently outperforming traditional solutions and advanced MLLMs.

Hierarchical Relation-augmented Representation Generalization for Few-shot Action Recognition

Few-shot action recognition (FSAR) aims to recognize novel action categories with few exemplars. Existing methods typically learn frame-level representations independently for each video by designing various inter-frame temporal modeling strategies. However, they neglect explicit relation modeling between videos and tasks, thus failing to capture shared temporal patterns across videos and reuse temporal knowledge from historical tasks. In light of this, we propose HR2G-shot, a Hierarchical Relation-augmented Representation Generalization framework for FSAR, which unifies three types of relation modeling (inter-frame, inter-video, and inter-task) to learn task-specific temporal patterns from a holistic view. In addition to conducting inter-frame temporal interactions, we further devise two components to respectively explore inter-video and inter-task relationships: i) Inter-video Semantic Correlation (ISC) performs cross-video frame-level interactions in a fine-grained manner, thereby capturing task-specific query features and learning intra- and inter-class temporal correlations among support features; ii) Inter-task Knowledge Transfer (IKT) retrieves and aggregates relevant temporal knowledge from the bank, which stores diverse temporal patterns from historical tasks. Extensive experiments on five benchmarks show that HR2G-shot outperforms current top-leading FSAR methods.

Efficient Token Compression for Vision Transformer with Spatial Information Preserved

Token compression is essential for reducing the computational and memory requirements of transformer models, enabling their deployment in resource-constrained environments. In this work, we propose an efficient and hardware-compatible token compression method called Prune and Merge. Our approach integrates token pruning and merging operations within transformer models to achieve layer-wise token compression. By introducing trainable merge and reconstruct matrices and utilizing shortcut connections, we efficiently merge tokens while preserving important information and enabling the restoration of pruned tokens. Additionally, we introduce a novel gradient-weighted attention scoring mechanism that computes token importance scores during the training phase, eliminating the need for separate computations during inference and enhancing compression efficiency. We also leverage gradient information to capture the global impact of tokens and automatically identify optimal compression structures. Extensive experiments on the ImageNet-1k and ADE20K datasets validate the effectiveness of our approach, achieving significant speed-ups with minimal accuracy degradation compared to state-of-the-art methods. For instance, on DeiT-Small, we achieve a 1.64$\times$ speed-up with only a 0.2\% drop in accuracy on ImageNet-1k. Moreover, by compressing segmenter models and comparing with existing methods, we demonstrate the superior performance of our approach in terms of efficiency and effectiveness. Code and models have been made available at https://github.com/NUST-Machine-Intelligence-Laboratory/prune_and_merge.

Twofold Debiasing Enhances Fine-Grained Learning with Coarse Labels

The Coarse-to-Fine Few-Shot (C2FS) task is designed to train models using only coarse labels, then leverages a limited number of subclass samples to achieve fine-grained recognition capabilities. This task presents two main challenges: coarse-grained supervised pre-training suppresses the extraction of critical fine-grained features for subcategory discrimination, and models suffer from overfitting due to biased distributions caused by limited fine-grained samples. In this paper, we propose the Twofold Debiasing (TFB) method, which addresses these challenges through detailed feature enhancement and distribution calibration. Specifically, we introduce a multi-layer feature fusion reconstruction module and an intermediate layer feature alignment module to combat the model's tendency to focus on simple predictive features directly related to coarse-grained supervision, while neglecting complex fine-grained level details. Furthermore, we mitigate the biased distributions learned by the fine-grained classifier using readily available coarse-grained sample embeddings enriched with fine-grained information. Extensive experiments conducted on five benchmark datasets demonstrate the efficacy of our approach, achieving state-of-the-art results that surpass competitive methods.

Explanatory Instructions: Towards Unified Vision Tasks Understanding and Zero-shot Generalization

Computer Vision (CV) has yet to fully achieve the zero-shot task generalization observed in Natural Language Processing (NLP), despite following many of the milestones established in NLP, such as large transformer models, extensive pre-training, and the auto-regression paradigm, among others. In this paper, we explore the idea that CV adopts discrete and terminological task definitions (\eg, ``image segmentation''), which may be a key barrier to zero-shot task generalization. Our hypothesis is that without truly understanding previously-seen tasks--due to these terminological definitions--deep models struggle to generalize to novel tasks. To verify this, we introduce Explanatory Instructions, which provide an intuitive way to define CV task objectives through detailed linguistic transformations from input images to outputs. We create a large-scale dataset comprising 12 million ``image input $\to$ explanatory instruction $\to$ output'' triplets, and train an auto-regressive-based vision-language model (AR-based VLM) that takes both images and explanatory instructions as input. By learning to follow these instructions, the AR-based VLM achieves instruction-level zero-shot capabilities for previously-seen tasks and demonstrates strong zero-shot generalization for unseen CV tasks. Code and dataset will be openly available on our GitHub repository.

The Key of Understanding Vision Tasks: Explanatory Instructions

Computer Vision (CV) has yet to fully achieve the zero-shot task generalization observed in Natural Language Processing (NLP), despite following many of the milestones established in NLP, such as large transformer models, extensive pre-training, and the auto-regression paradigm, among others. In this paper, we explore the idea that CV adopts discrete and terminological task definitions (\eg, ``image segmentation''), which may be a key barrier to zero-shot task generalization. Our hypothesis is that without truly understanding previously-seen tasks--due to these terminological definitions--deep models struggle to generalize to novel tasks. To verify this, we introduce Explanatory Instructions, which provide an intuitive way to define CV task objectives through detailed linguistic transformations from input images to outputs. We create a large-scale dataset comprising 12 million ``image input $\to$ explanatory instruction $\to$ output'' triplets, and train an auto-regressive-based vision-language model (AR-based VLM) that takes both images and explanatory instructions as input. By learning to follow these instructions, the AR-based VLM achieves instruction-level zero-shot capabilities for previously-seen tasks and demonstrates strong zero-shot generalization for unseen CV tasks. Code and dataset will be openly available on our GitHub repository.

FTMoMamba: Motion Generation with Frequency and Text State Space Models

Diffusion models achieve impressive performance in human motion generation. However, current approaches typically ignore the significance of frequency-domain information in capturing fine-grained motions within the latent space (e.g., low frequencies correlate with static poses, and high frequencies align with fine-grained motions). Additionally, there is a semantic discrepancy between text and motion, leading to inconsistency between the generated motions and the text descriptions. In this work, we propose a novel diffusion-based FTMoMamba framework equipped with a Frequency State Space Model (FreqSSM) and a Text State Space Model (TextSSM). Specifically, to learn fine-grained representation, FreqSSM decomposes sequences into low-frequency and high-frequency components, guiding the generation of static pose (e.g., sits, lay) and fine-grained motions (e.g., transition, stumble), respectively. To ensure the consistency between text and motion, TextSSM encodes text features at the sentence level, aligning textual semantics with sequential features. Extensive experiments show that FTMoMamba achieves superior performance on the text-to-motion generation task, especially gaining the lowest FID of 0.181 (rather lower than 0.421 of MLD) on the HumanML3D dataset.

UnitedVLN: Generalizable Gaussian Splatting for Continuous Vision-Language Navigation

Vision-and-Language Navigation (VLN), where an agent follows instructions to reach a target destination, has recently seen significant advancements. In contrast to navigation in discrete environments with predefined trajectories, VLN in Continuous Environments (VLN-CE) presents greater challenges, as the agent is free to navigate any unobstructed location and is more vulnerable to visual occlusions or blind spots. Recent approaches have attempted to address this by imagining future environments, either through predicted future visual images or semantic features, rather than relying solely on current observations. However, these RGB-based and feature-based methods lack intuitive appearance-level information or high-level semantic complexity crucial for effective navigation. To overcome these limitations, we introduce a novel, generalizable 3DGS-based pre-training paradigm, called UnitedVLN, which enables agents to better explore future environments by unitedly rendering high-fidelity 360 visual images and semantic features. UnitedVLN employs two key schemes: search-then-query sampling and separate-then-united rendering, which facilitate efficient exploitation of neural primitives, helping to integrate both appearance and semantic information for more robust navigation. Extensive experiments demonstrate that UnitedVLN outperforms state-of-the-art methods on existing VLN-CE benchmarks.