Object detection methods have evolved from closed-set to open-set paradigms over the years. Current open-set object detectors, however, remain constrained by their exclusive reliance on positive indicators based on given prompts like text descriptions or visual exemplars. This positive-only paradigm experiences consistent vulnerability to visually similar but semantically different distractors. We propose T-Rex-Omni, a novel framework that addresses this limitation by incorporating negative visual prompts to negate hard negative distractors. Specifically, we first introduce a unified visual prompt encoder that jointly processes positive and negative visual prompts. Next, a training-free Negating Negative Computing (NNC) module is proposed to dynamically suppress negative responses during the probability computing stage. To further boost performance through fine-tuning, our Negating Negative Hinge (NNH) loss enforces discriminative margins between positive and negative embeddings. T-Rex-Omni supports flexible deployment in both positive-only and joint positive-negative inference modes, accommodating either user-specified or automatically generated negative examples. Extensive experiments demonstrate remarkable zero-shot detection performance, significantly narrowing the performance gap between visual-prompted and text-prompted methods while showing particular strength in long-tailed scenarios (51.2 AP_r on LVIS-minival). This work establishes negative prompts as a crucial new dimension for advancing open-set visual recognition systems.

Chenfei Liao, Wensong Wang, Zichen Wen, Xu Zheng, Yiyu Wang, Haocong He, Yuanhuiyi Lyu, Lutao Jiang, Xin Zou, Yuqian Fu, Bin Ren, Linfeng Zhang, Xuming Hu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Diffusion Transformers (DiT) have revolutionized high-fidelity image and video synthesis, yet their computational demands remain prohibitive for real-time applications. To solve this problem, feature caching has been proposed to accelerate diffusion models by caching the features in the previous timesteps and then reusing them in the following timesteps. However, at timesteps with significant intervals, the feature similarity in diffusion models decreases substantially, leading to a pronounced increase in errors introduced by feature caching, significantly harming the generation quality. To solve this problem, we propose TaylorSeer, which firstly shows that features of diffusion models at future timesteps can be predicted based on their values at previous timesteps. Based on the fact that features change slowly and continuously across timesteps, TaylorSeer employs a differential method to approximate the higher-order derivatives of features and predict features in future timesteps with Taylor series expansion. Extensive experiments demonstrate its significant effectiveness in both image and video synthesis, especially in high acceleration ratios. For instance, it achieves an almost lossless acceleration of 4.99$\times$ on FLUX and 5.00$\times$ on HunyuanVideo without additional training. On DiT, it achieves $3.41$ lower FID compared with previous SOTA at $4.53$$\times$ acceleration. %Our code is provided in the supplementary materials and will be made publicly available on GitHub. Our codes have been released in Github:https://github.com/Shenyi-Z/TaylorSeer

Fusing and balancing multi-modal inputs from novel sensors for dense prediction tasks, particularly semantic segmentation, is critically important yet remains a significant challenge. One major limitation is the tendency of multi-modal frameworks to over-rely on easily learnable modalities, a phenomenon referred to as unimodal dominance or bias. This issue becomes especially problematic in real-world scenarios where the dominant modality may be unavailable, resulting in severe performance degradation. To this end, we apply a simple but effective plug-and-play regularization term based on functional entropy, which introduces no additional parameters or modules. This term is designed to intuitively balance the contribution of each visual modality to the segmentation results. Specifically, we leverage the log-Sobolev inequality to bound functional entropy using functional-Fisher-information. By maximizing the information contributed by each visual modality, our approach mitigates unimodal dominance and establishes a more balanced and robust segmentation framework. A multi-scale regularization module is proposed to apply our proposed plug-and-play term on high-level features and also segmentation predictions for more balanced multi-modal learning. Extensive experiments on three datasets demonstrate that our proposed method achieves superior performance, i.e., +13.94%, +3.25%, and +3.64%, without introducing any additional parameters.

Despite their impressive capabilities, Multimodal Large Language Models (MLLMs) are prone to hallucinations, i.e., the generated content that is nonsensical or unfaithful to input sources. Unlike in LLMs, hallucinations in MLLMs often stem from the sensitivity of text decoder to visual tokens, leading to a phenomenon akin to "amnesia" about visual information. To address this issue, we propose MemVR, a novel decoding paradigm inspired by common cognition: when the memory of an image seen the moment before is forgotten, people will look at it again for factual answers. Following this principle, we treat visual tokens as supplementary evidence, re-injecting them into the MLLM through Feed Forward Network (FFN) as “key-value memory” at the middle trigger layer. This look-twice mechanism occurs when the model exhibits high uncertainty during inference, effectively enhancing factual alignment. Comprehensive experimental evaluations demonstrate that MemVR significantly mitigates hallucination across various MLLMs and excels in general benchmarks without incurring additional time overhead.

Recent text-to-image generative models, e.g., Stable Diffusion V3 and Flux, have achieved notable progress. However, these models are strongly restricted to their limited knowledge, a.k.a., their own fixed parameters, that are trained with closed datasets. This leads to significant hallucinations or distortions when facing fine-grained and unseen novel real-world objects, e.g., the appearance of the Tesla Cybertruck. To this end, we present the first real-object-based retrieval-augmented generation framework (RealRAG), which augments fine-grained and unseen novel object generation by learning and retrieving real-world images to overcome the knowledge gaps of generative models. Specifically, to integrate missing memory for unseen novel object generation, we train a reflective retriever by self-reflective contrastive learning, which injects the generator’s knowledge into the sef-reflective negatives, ensuring that the retrieved augmented images compensate for the model’s missing knowledge. Furthermore, the real-object-based framework integrates fine-grained visual knowledge for the generative models, tackling the distortion problem and improving the realism for fine-grained object generation. Our Real-RAG is superior in its modular application to all types of state-of-the-art text-to-image generative models and also delivers remarkable performance boosts with all of them, such as a gain of 16.18% FID score with the auto-regressive model on the Stanford Car benchmark.

Diffusion models have revolutionized high-fidelity image and video synthesis, yet their computational demands remain prohibitive for real-time applications. These models face two fundamental challenges: strict temporal dependencies preventing parallelization, and computationally intensive forward passes required at each denoising step. Drawing inspiration from speculative decoding in large language models, we present SpeCa, a novel 'Forecast-then-verify' acceleration framework that effectively addresses both limitations. SpeCa's core innovation lies in introducing Speculative Sampling to diffusion models, predicting intermediate features for subsequent timesteps based on fully computed reference timesteps. Our approach implements a parameter-free verification mechanism that efficiently evaluates prediction reliability, enabling real-time decisions to accept or reject each prediction while incurring negligible computational overhead. Furthermore, SpeCa introduces sample-adaptive computation allocation that dynamically modulates resources based on generation complexity, allocating reduced computation for simpler samples while preserving intensive processing for complex instances. Experiments demonstrate 6.34x acceleration on FLUX with minimal quality degradation (5.5% drop), 7.3x speedup on DiT while preserving generation fidelity, and 79.84% VBench score at 6.1x acceleration for HunyuanVideo. The verification mechanism incurs minimal overhead (1.67%-3.5% of full inference costs), establishing a new paradigm for efficient diffusion model inference while maintaining generation quality even at aggressive acceleration ratios. Our codes have been released in Github: \textbf{https://github.com/Shenyi-Z/Cache4Diffusion}

360-degree monocular depth estimation plays a crucial role in scene understanding owing to its 180-degree by 360-degree field-of-view (FoV). To mitigate the distortions brought by equirectangular projection, existing methods typically divide 360-degree images into distortion-less perspective patches. However, since these patches are processed independently, depth inconsistencies are often introduced due to scale drift among patches. Recently, video depth estimation (VDE) models have leveraged temporal consistency for stable depth predictions across frames. Inspired by this, we propose to represent a 360-degree image as a sequence of perspective frames, mimicking the viewpoint adjustments users make when exploring a 360-degree scenario in virtual reality. Thus, the spatial consistency among perspective depth patches can be enhanced by exploiting the temporal consistency inherent in VDE models. To this end, we introduce a training-free pipeline for 360-degree monocular depth estimation, called ST²360D. Specifically, ST²360D transforms a 360-degree image into perspective video frames, predicts video depth maps using VDE models, and seamlessly merges these predictions into a complete 360-degree depth map. To generate sequenced perspective frames that align with VDE models, we propose two tailored strategies. First, a spherical-uniform sampling (SUS) strategy is proposed to facilitate uniform sampling of perspective views across the sphere, avoiding oversampling in polar regions typically with limited structural details. Second, a latitude-guided scanning (LGS) strategy is introduced to organize the frames into a coherent sequence, starting from the equator, prioritizing low-latitude slices, and progressively moving toward higher latitudes. Extensive experiments demonstrate that ST²360D achieves strong zero-shot capability on several datasets, supporting resolutions up to 4K.

Despite their powerful capabilities, multimodal large language models (MLLMs) suffer from considerable computational overhead due to their reliance on massive visual tokens. Recent studies have explored token pruning to alleviate this problem, which typically uses text-vision cross-attention or [CLS] attention to assess and discard redundant visual tokens. In this work, we identify a critical limitation of such attention-first pruning approaches, i.e., they tend to preserve semantically similar tokens, resulting in pronounced performance drops under high pruning rates. To this end, we propose HoloV, a simple yet effective, plug-and-play visual token pruning framework for efficient inference. Distinct from previous attention-first schemes, HoloV rethinks token retention from a holistic perspective. By adaptively distributing the pruning budget across different spatial crops, HoloV ensures that the retained tokens capture the global visual context rather than isolated salient features. This strategy minimizes representational collapse and maintains task-relevant information even under aggressive pruning. Experimental results demonstrate that our HoloV achieves superior performance across various tasks, MLLM architectures, and pruning ratios compared to SOTA methods. For instance, LLaVA1.5 equipped with HoloV preserves 95.8% of the original performance after pruning 88.9% of visual tokens, achieving superior efficiency-accuracy trade-offs.

We present UniBind, a flexible and efficient approach that learns a unified representation space for seven diverse modalities - image, text, audio, point cloud, thermal, video, and event data. Existing works, e.g., ImageBind [13], treat the image as the central modality and build an image-centered representation space; however, the space may be sub-optimal as it leads to an unbalanced representation space among all modalities. Moreover, the category names are directly used to extract text embeddings for the down-stream tasks, making it hardly possible to represent the se-mantics of multi-modal data. The ‘out-of-the-box’ insight of our UniBind is to make the alignment centers modality-agnostic and further learn a unified and balanced repre-sentation space, empowered by the large language mod-els (LLMs). UniBind is superior in its flexible application to all CLIP-style models and delivers remarkable performance boosts. To make this possible, we 1) construct a knowledge base of text with the help of LLMs and multi-modal LLMs; 2) adaptively build LLM-augmented class-wise embedding centers on top of the knowledge base and encoded visual embeddings; 3) align all the embeddings to the LLM-augmented embedding centers via contrastive learning to achieve a unified and balanced representation space. UniBind shows strong zero-shot recognition performance gains over prior arts by an average of 6.36%. Fi-nally, we achieve new state-of-the-art performance, e.g., a 6.75% gain on ImageNet, on the multi-modal fine-tuning setting while reducing 90% of the learnable parameters.

Event cameras have recently been shown beneficial for practical vision tasks, such as action recognition, thanks to their high temporal resolution, power efficiency, and reduced privacy concerns. However, current research is hindered by 1) the difficulty in processing events because of their prolonged duration and dynamic actions with complex and ambiguous semantics and 2) the redundant action depiction of the event frame representation with fixed stacks. We find language naturally conveys abundant semantic information, rendering it stunningly superior in reducing semantic uncertainty. In light of this, we propose ExACT, a novel approach that, for the first time, tackles event-based action recognition from a cross-modal conceptualizing perspective. Our ExACT brings two technical contributions. Firstly, we propose an adaptive fine-grained event (AFE) representation to adaptively filter out the repeated events for the stationary objects while preserving dynamic ones. This subtly enhances the performance of ExACT without extra computational cost. Then, we propose a conceptual reasoning-based uncertainty estimation module, which simulates the recognition process to enrich the semantic representation. In particular, conceptual reasoning builds the temporal relation based on the action semantics, and uncertainty estimation tackles the semantic uncertainty of actions based on the distributional representation. Experiments show that our ExACT achieves superior recognition accuracy of 94.83%(+2.23%), 90.10%(+37.47%) and 67.24% on PAF, HARDVS and our SeAct datasets respectively.

The ability to detect objects in all lighting (i.e., normal-, over-, and under-exposed) conditions is crucial for real-world applications, such as self-driving. Traditional RGB-based detectors often fail under such varying lighting conditions. Therefore, recent works utilize novel event cameras to supplement or guide the RGB modality; however, these methods typically adopt asymmetric network structures that rely predominantly on the RGB modality, resulting in limited robustness for all-day detection. In this paper, we propose EOLO, a novel object detection framework that achieves robust and efficient all-day detection by fusing both RGB and event modalities. Our EOLO framework is built based on a lightweight spiking neural network (SNN) to efficiently leverage the asynchronous property of events. Buttressed by it, we first introduce an Event Temporal Attention (ETA) module to learn the high temporal information from events while preserving crucial edge information. Secondly, as different modalities exhibit varying levels of importance under diverse lighting conditions, we propose a novel Symmetric RGB-Event Fusion (SREF) module to effectively fuse RGB-Event features without relying on a specific modality, thus ensuring a balanced and adaptive fusion for all-day detection. In addition, to compensate for the lack of paired RGB-Event datasets for all-day training and evaluation, we propose an event synthesis approach based on the randomized optical flow that allows for directly generating the event frame from a single exposure image. We further build two new datasets, E-MSCOCO and E-VOC based on the popular benchmarks MSCOCO and PASCAL VOC. Extensive experiments demonstrate that our EOLO outperforms the state-of-the-art detectors, e.g., RENet [1], by a substantial margin (+3.74% mAP50) in all lighting conditions. Our code and datasets will be available at https://vlislab22.github.io/EOLO/.