Hybrid-Level Instruction Injection for Video Token Compression in Multi-modal Large Language Models

Recent Multi-modal Large Language Models (MLLMs) have been challenged by the computational overhead resulting from massive video frames, often alleviated through compression strategies. However, the visual content is not equally contributed to user instructions, existing strategies (\eg, average pool) inevitably lead to the loss of potentially useful information. To tackle this, we propose the Hybrid-level Instruction Injection Strategy for Conditional Token Compression in MLLMs (HICom), utilizing the instruction as a condition to guide the compression from both local and global levels. This encourages the compression to retain the maximum amount of user-focused information while reducing visual tokens to minimize computational burden. Specifically, the instruction condition is injected into the grouped visual tokens at the local level and the learnable tokens at the global level, and we conduct the attention mechanism to complete the conditional compression. From the hybrid-level compression, the instruction-relevant visual parts are highlighted while the temporal-spatial structure is also preserved for easier understanding of LLMs. To further unleash the potential of HICom, we introduce a new conditional pre-training stage with our proposed dataset HICom-248K. Experiments show that our HICom can obtain distinguished video understanding ability with fewer tokens, increasing the performance by 2.43\% average on three multiple-choice QA benchmarks and saving 78.8\% tokens compared with the SOTA method. The code is available at https://github.com/lntzm/HICom.

Rethinking Video Tokenization: A Conditioned Diffusion-based Approach

Video tokenizers, which transform videos into compact latent representations, are key to video generation. Existing video tokenizers are based on the VAE architecture and follow a paradigm where an encoder compresses videos into compact latents, and a deterministic decoder reconstructs the original videos from these latents. In this paper, we propose a novel \underline{\textbf{C}}onditioned \underline{\textbf{D}}iffusion-based video \underline{\textbf{T}}okenizer entitled \textbf{\ourmethod}, which departs from previous methods by replacing the deterministic decoder with a 3D causal diffusion model. The reverse diffusion generative process of the decoder is conditioned on the latent representations derived via the encoder. With a feature caching and sampling acceleration, the framework efficiently reconstructs high-fidelity videos of arbitrary lengths. Results show that {\ourmethod} achieves state-of-the-art performance in video reconstruction tasks using just a single-step sampling. Even a smaller version of {\ourmethod} still achieves reconstruction results on par with the top two baselines. Furthermore, the latent video generation model trained using {\ourmethod} also shows superior performance.

UFO: A Unified Approach to Fine-grained Visual Perception via Open-ended Language Interface

Generalist models have achieved remarkable success in both language and vision-language tasks, showcasing the potential of unified modeling. However, effectively integrating fine-grained perception tasks like detection and segmentation into these models remains a significant challenge. This is primarily because these tasks often rely heavily on task-specific designs and architectures that can complicate the modeling process. To address this challenge, we present \ours, a framework that \textbf{U}nifies \textbf{F}ine-grained visual perception tasks through an \textbf{O}pen-ended language interface. By transforming all perception targets into the language space, \ours unifies object-level detection, pixel-level segmentation, and image-level vision-language tasks into a single model. Additionally, we introduce a novel embedding retrieval approach that relies solely on the language interface to support segmentation tasks. Our framework bridges the gap between fine-grained perception and vision-language tasks, significantly simplifying architectural design and training strategies while achieving comparable or superior performance to methods with intricate task-specific designs. After multi-task training on five standard visual perception datasets, \ours outperforms the previous state-of-the-art generalist models by 12.3 mAP on COCO instance segmentation and 3.3 mIoU on ADE20K semantic segmentation. Furthermore, our method seamlessly integrates with existing MLLMs, effectively combining fine-grained perception capabilities with their advanced language abilities, thereby enabling more challenging tasks such as reasoning segmentation. Code and models are available at https://github.com/nnnth/UFO.

What Is a Good Caption? A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Coverage of MLLMs

Recent advancements in Multimodal Large Language Models (MLLMs) have rendered traditional visual captioning benchmarks obsolete, as they primarily evaluate short descriptions with outdated metrics. While recent benchmarks address these limitations by decomposing captions into visual elements and adopting model-based evaluation, they remain incomplete-overlooking critical aspects, while providing vague, non-explanatory scores. To bridge this gap, we propose CV-CapBench, a Comprehensive Visual Caption Benchmark that systematically evaluates caption quality across 6 views and 13 dimensions. CV-CapBench introduces precision, recall, and hit rate metrics for each dimension, uniquely assessing both correctness and coverage. Experiments on leading MLLMs reveal significant capability gaps, particularly in dynamic and knowledge-intensive dimensions. These findings provide actionable insights for future research. The code and data will be released.

AlignZeg: Mitigating Objective Misalignment for Zero-shot Semantic Segmentation

A serious issue that harms the performance of zero-shot visual recognition is named objective misalignment, i.e., the learning objective prioritizes improving the recognition accuracy of seen classes rather than unseen classes, while the latter is the true target to pursue. This issue becomes more significant in zero-shot image segmentation because the stronger (i.e., pixel-level) supervision brings a larger gap between seen and unseen classes. To mitigate it, we propose a novel architecture named AlignZeg, which embodies a comprehensive improvement of the segmentation pipeline, including proposal extraction, classification, and correction, to better fit the goal of zero-shot segmentation. (1) Mutually-Refined Proposal Extraction. AlignZeg harnesses a mutual interaction between mask queries and visual features, facilitating detailed class-agnostic mask proposal extraction. (2) Generalization-Enhanced Proposal Classification. AlignZeg introduces synthetic data and incorporates multiple background prototypes to allocate a more generalizable feature space. (3) Predictive Bias Correction. During the inference stage, AlignZeg uses a class indicator to find potential unseen class proposals followed by a prediction postprocess to correct the prediction bias. Experiments demonstrate that AlignZeg markedly enhances zero-shot semantic segmentation, as shown by an average 3.8% increase in hIoU, primarily attributed to a 7.1% improvement in identifying unseen classes, and we further validate that the improvement comes from alleviating the objective misalignment issue.

Towards Balanced Alignment: Modal-Enhanced Semantic Modeling for Video Moment Retrieval

Video Moment Retrieval (VMR) aims to retrieve temporal segments in untrimmed videos corresponding to a given language query by constructing cross-modal alignment strategies. However, these existing strategies are often sub-optimal since they ignore the modality imbalance problem, \textit{i.e.}, the semantic richness inherent in videos far exceeds that of a given limited-length sentence. Therefore, in pursuit of better alignment, a natural idea is enhancing the video modality to filter out query-irrelevant semantics, and enhancing the text modality to capture more segment-relevant knowledge. In this paper, we introduce Modal-Enhanced Semantic Modeling (MESM), a novel framework for more balanced alignment through enhancing features at two levels. First, we enhance the video modality at the frame-word level through word reconstruction. This strategy emphasizes the portions associated with query words in frame-level features while suppressing irrelevant parts. Therefore, the enhanced video contains less redundant semantics and is more balanced with the textual modality. Second, we enhance the textual modality at the segment-sentence level by learning complementary knowledge from context sentences and ground-truth segments. With the knowledge added to the query, the textual modality thus maintains more meaningful semantics and is more balanced with the video modality. By implementing two levels of MESM, the semantic information from both modalities is more balanced to align, thereby bridging the modality gap. Experiments on three widely used benchmarks, including the out-of-distribution settings, show that the proposed framework achieves a new start-of-the-art performance with notable generalization ability (e.g., 4.42% and 7.69% average gains of R1@0.7 on Charades-STA and Charades-CG). The code will be available at https://github.com/lntzm/MESM.

Dual-Stream Knowledge-Preserving Hashing for Unsupervised Video Retrieval

Unsupervised video hashing usually optimizes binary codes by learning to reconstruct input videos. Such reconstruction constraint spends much effort on frame-level temporal context changes without focusing on video-level global semantics that are more useful for retrieval. Hence, we address this problem by decomposing video information into reconstruction-dependent and semantic-dependent information, which disentangles the semantic extraction from reconstruction constraint. Specifically, we first design a simple dual-stream structure, including a temporal layer and a hash layer. Then, with the help of semantic similarity knowledge obtained from self-supervision, the hash layer learns to capture information for semantic retrieval, while the temporal layer learns to capture the information for reconstruction. In this way, the model naturally preserves the disentangled semantics into binary codes. Validated by comprehensive experiments, our method consistently outperforms the state-of-the-arts on three video benchmarks.

Balanced Classification: A Unified Framework for Long-Tailed Object Detection

Conventional detectors suffer from performance degradation when dealing with long-tailed data due to a classification bias towards the majority head categories. In this paper, we contend that the learning bias originates from two factors: 1) the unequal competition arising from the imbalanced distribution of foreground categories, and 2) the lack of sample diversity in tail categories. To tackle these issues, we introduce a unified framework called BAlanced CLassification (BACL), which enables adaptive rectification of inequalities caused by disparities in category distribution and dynamic intensification of sample diversities in a synchronized manner. Specifically, a novel foreground classification balance loss (FCBL) is developed to ameliorate the domination of head categories and shift attention to difficult-to-differentiate categories by introducing pairwise class-aware margins and auto-adjusted weight terms, respectively. This loss prevents the over-suppression of tail categories in the context of unequal competition. Moreover, we propose a dynamic feature hallucination module (FHM), which enhances the representation of tail categories in the feature space by synthesizing hallucinated samples to introduce additional data variances. In this divide-and-conquer approach, BACL sets a new state-of-the-art on the challenging LVIS benchmark with a decoupled training pipeline, surpassing vanilla Faster R-CNN with ResNet-50-FPN by 5.8% AP and 16.1% AP for overall and tail categories. Extensive experiments demonstrate that BACL consistently achieves performance improvements across various datasets with different backbones and architectures. Code and models are available at https://github.com/Tianhao-Qi/BACL.

MomentDiff: Generative Video Moment Retrieval from Random to Real

Video moment retrieval pursues an efficient and generalized solution to identify the specific temporal segments within an untrimmed video that correspond to a given language description. To achieve this goal, we provide a generative diffusion-based framework called MomentDiff, which simulates a typical human retrieval process from random browsing to gradual localization. Specifically, we first diffuse the real span to random noise, and learn to denoise the random noise to the original span with the guidance of similarity between text and video. This allows the model to learn a mapping from arbitrary random locations to real moments, enabling the ability to locate segments from random initialization. Once trained, MomentDiff could sample random temporal segments as initial guesses and iteratively refine them to generate an accurate temporal boundary. Different from discriminative works (e.g., based on learnable proposals or queries), MomentDiff with random initialized spans could resist the temporal location biases from datasets. To evaluate the influence of the temporal location biases, we propose two anti-bias datasets with location distribution shifts, named Charades-STA-Len and Charades-STA-Mom. The experimental results demonstrate that our efficient framework consistently outperforms state-of-the-art methods on three public benchmarks, and exhibits better generalization and robustness on the proposed anti-bias datasets. The code, model, and anti-bias evaluation datasets are available at https://github.com/IMCCretrieval/MomentDiff.