Audio-Visual Learning

Language models have been effectively applied to modeling natural signals, such as images, video, speech, and audio. A crucial component of these models is the codec tokenizer, which compresses high-dimensional natural signals into lower-dimensional discrete tokens. In this paper, we introduce WavTokenizer, which offers several advantages over previous SOTA acoustic codec models in the audio domain: 1) extreme compression. By compressing the layers of quantizers and the temporal dimension of the discrete codec, one-second audio of 24kHz sampling rate requires only a single quantizer with 40 or 75 tokens. 2) improved subjective quality. Despite the reduced number of tokens, WavTokenizer achieves state-of-the-art reconstruction quality with outstanding UTMOS scores and inherently contains richer semantic information. Specifically, we achieve these results by designing a broader VQ space, extended contextual windows, and improved attention networks, as well as introducing a powerful multi-scale discriminator and an inverse Fourier transform structure. We conducted extensive reconstruction experiments in the domains of speech, audio, and music. WavTokenizer exhibited strong performance across various objective and subjective metrics compared to state-of-the-art models. We also tested semantic information, VQ utilization, and adaptability to generative models. Comprehensive ablation studies confirm the necessity of each module in WavTokenizer. The code is available at https://github.com/jishengpeng/WavTokenizer.

Visual imitation learning methods demonstrate strong performance, yet they lack generalization when faced with visual input perturbations like variations in lighting and textures. This limitation hampers their practical application in real-world settings. To address this, we propose ***Stem-OB*** that leverages the inversion process of pretrained image diffusion models to suppress low-level visual differences while maintaining high-level scene structures. This image inversion process is akin to transforming the observation into a shared representation, from which other observations also stem. *Stem-OB* offers a simple yet effective plug-and-play solution that stands in contrast to data augmentation approaches. It demonstrates robustness to various unspecified appearance changes without the need for additional training. We provide theoretical insights and empirical results that validate the efficacy of our approach in simulated and real settings. *Stem-OB* shows an exceptionally significant improvement in real-world robotic tasks, where challenging light and appearance changes are present, with an average increase of **22.2%** in success rates compared to the best baseline. Please refer to [this link](https://stem-ob.github.io/) for more videos and details.

Long-context capability is critical for multi-modal foundation models, especially for long video understanding. We introduce LongVILA, a full-stack solution for long-context visual-language models by co-designing the algorithm and system. For model training, we upgrade existing VLMs to support long video understanding by incorporating two additional stages, i.e., long context extension and long video supervised fine-tuning. However, training on long video is computationally and memory intensive. We introduce the long-context Multi-Modal Sequence Parallelism (MM-SP) system that efficiently parallelizes long video training and inference, enabling 2M context length training on 256 GPUs without any gradient checkpointing. LongVILA efficiently extends the number of video frames of VILA from 8 to 2048, achieving 99.8% accuracy in 6,000-frame (more than 1 million tokens) video needle-in-a-haystack. LongVILA-7B demonstrates strong accuracy on 9 popular video benchmarks, e.g., 65.1% VideoMME with subtitle. Besides, MM-SP is 2.1x - 5.7x faster than ring style sequence parallelism and 1.1x - 1.4x faster than Megatron with a hybrid context and tensor parallelism. Moreover, it seamlessly integrates with Hugging Face Transformers.

This paper investigates visual analogical reasoning in large multimodal models (LMMs) compared to human adults and children. A “visual analogy” is an abstract rule inferred from one image and applied to another. While benchmarks exist for testing visual reasoning in LMMs, they require advanced skills and omit basic visual analogies that even young children can make. Inspired by developmental psychology, we propose a new benchmark of 4,300 visual transformations of everyday objects to test LMMs on visual analogical reasoning and compare them to children (ages three to five) and to adults. We structure the evaluation into three stages: identifying what changed (e.g., color, number, etc.), how it changed (e.g., added one object), and applying the rule to new scenarios. Our findings show that while GPT-o1, GPT-4V, LLaVA-1.5, and MANTIS identify the “what” effectively, they struggle with quantifying the “how” and extrapolating this rule to new objects. In contrast, children and adults exhibit much stronger analogical reasoning at all three stages. Additionally, the strongest tested model, GPT-o1, performs better in tasks involving simple surface-level visual attributes like color and size, correlating with quicker human adult response times. Conversely, more complex tasks such as number, rotation, and reflection, which necessitate extensive cognitive processing and understanding of extrinsic spatial properties in the physical world, present more significant challenges. Altogether, these findings highlight the limitations of training models on data that primarily consists of 2D images and text.

Decomposing geometry, materials and lighting from a set of images, namely inverse rendering, has been a long-standing problem in computer vision and graphics. Recent advances in neural rendering enable photo-realistic and plausible inverse rendering results. The emergence of 3D Gaussian Splatting has boosted it to the next level by showing real-time rendering potentials. An intuitive finding is that the models used for inverse rendering do not take into account the dependency of opacity w.r.t. material properties, namely cross section, as suggested by optics. Therefore, we develop a novel approach that adds this dependency to the modeling itself. Inspired by radiative transfer, we augment the opacity term by introducing a neural network that takes as input material properties to provide modeling of cross section and a physically correct activation function. The gradients for material properties are therefore not only from color but also from opacity, facilitating a constraint for their optimization. Therefore, the proposed method incorporates more accurate physical properties compared to previous works. We implement our method into 3 different baselines that use Gaussian Splatting for inverse rendering and achieve significant improvements universally in terms of novel view synthesis and material modeling.

Hierarchical visual processing is essential for cognitive functions like object recognition and spatial localization. Traditional studies of the neural basis of these computations have focused on single-neuron activity, but recent advances in large-scale neural recordings emphasize the growing need to understand computations at the population level. Digital twins-computational models trained on neural data-have successfully replicated single-neuron behavior, but their effectiveness in capturing the joint activity of neurons remains unclear. In this study, we investigate how well digital twins describe population responses in mouse visual cortex. We show that these models fail to accurately represent the geometry of population activity, particularly its differentiability and how this geometry evolves across the visual hierarchy. To address this, we explore how dataset, network architecture, loss function, and training method affect the ability of digital twins to recapitulate population properties. We demonstrate that improving model alignment with experiments requires training strategies that enhance robustness and generalization, reflecting principles observed in biological systems. These findings underscore the need to evaluate digital twins from multiple perspectives, identify key areas for refinement, and establish a foundation for using these models to explore neural computations at the population level.

Following the success of Large Language Models (LLMs), expanding their boundaries to new modalities represents a significant paradigm shift in multimodal understanding. Human perception is inherently multimodal, relying not only on text but also on auditory and visual cues for a complete understanding of the world. In recognition of this fact, audio-visual LLMs have recently emerged. Despite promising developments, the lack of dedicated benchmarks poses challenges for understanding and evaluating models. In this work, we show that audio-visual LLMs struggle to discern subtle relationships between audio and visual signals, leading to hallucinations and highlighting the need for reliable benchmarks. To address this, we introduce AVHBench, the first comprehensive benchmark specifically designed to evaluate the perception and comprehension capabilities of audio-visual LLMs. Our benchmark includes tests for assessing hallucinations, as well as the cross-modal matching and reasoning abilities of these models. Our results reveal that most existing audio-visual LLMs struggle with hallucinations caused by cross-interactions between modalities, due to their limited capacity to perceive complex multimodal signals and their relationships. Additionally, we demonstrate that simple training with our AVHBench improves robustness of audio-visual LLMs against hallucinations. Dataset: https://github.com/kaist-ami/AVHBench

This paper aims to address the challenge of hallucinations in Multimodal Large Language Models (MLLMs) particularly for dense image captioning tasks. To tackle the challenge, we identify the current lack of a metric that finely measures the caption quality in concept level. We hereby introduce HalFscore, a novel metric built upon the language graph and is designed to evaluate both the accuracy and completeness of dense captions at a granular level. Additionally, we identify the root cause of hallucination as the model's over-reliance on its language prior. To address this, we propose PerturboLLaVA, which reduces the model's reliance on the language prior by incorporating adversarially perturbed text during training. This method enhances the model's focus on visual inputs, effectively reducing hallucinations and producing accurate, image-grounded descriptions without incurring additional computational overhead. PerturboLLaVA significantly improves the fidelity of generated captions, outperforming existing approaches in handling multimodal hallucinations and achieving improved performance across general multimodal benchmarks.

How do two deep neural networks differ in how they arrive at a decision? Measuring the similarity of deep networks has been a long-standing open question. Most existing methods provide a single number to measure the similarity of two networks at a given layer, but give no insight into what makes them similar or dissimilar. We introduce an interpretable representational similarity method (RSVC) to compare two networks. We use RSVC to discover shared and unique visual concepts between two models. We show that some aspects of model differences can be attributed to unique concepts discovered by one model that are not well represented in the other. Finally, we conduct extensive evaluation across different vision model architectures and training protocols to demonstrate its effectiveness.

While audio-visual learning equips models with a richer understanding of the real world by leveraging multiple sensory modalities, this integration also introduces new vulnerabilities to adversarial attacks. In this paper, we present a comprehensive study of the adversarial robustness of audio-visual models, considering both temporal and modality-specific vulnerabilities. We propose two powerful adversarial attacks: 1) a temporal invariance attack that exploits the inherent temporal redundancy across consecutive time segments and 2) a modality misalignment attack that introduces incongruence between the audio and visual modalities. These attacks are designed to thoroughly assess the robustness of audio-visual models against diverse threats. Furthermore, to defend against such attacks, we introduce a novel audio-visual adversarial training framework. This framework addresses key challenges in vanilla adversarial training by incorporating efficient adversarial perturbation crafting tailored to multi-modal data and an adversarial curriculum strategy. Extensive experiments in the Kinetics-Sounds dataset demonstrate that our proposed temporal and modality-based attacks in degrading model performance can achieve state-of-the-art performance, while our adversarial training defense largely improves the adversarial robustness as well as the adversarial training efficiency.

Joint-embedding predictive architectures (JEPAs) have shown substantial promise in self-supervised representation learning, yet their application in generative modeling remains underexplored. Conversely, diffusion models have demonstrated significant efficacy in modeling arbitrary probability distributions. In this paper, we introduce Denoising with a Joint-Embedding Predictive Architecture (D-JEPA), pioneering the integration of JEPA within generative modeling. By recognizing JEPA as a form of masked image modeling, we reinterpret it as a generalized next-token prediction strategy, facilitating data generation in an auto-regressive manner. Furthermore, we incorporate diffusion loss to model the per-token probability distribution, enabling data generation in a continuous space. We also adapt flow matching loss as an alternative to diffusion loss, thereby enhancing the flexibility of D-JEPA. Empirically, with increased GFLOPs, D-JEPA consistently achieves lower FID scores with fewer training epochs, indicating its good scalability. Our base, large, and huge models outperform all previous generative models across all scales on ImageNet conditional generation benchmarks. Beyond image generation, D-JEPA is well-suited for other continuous data modeling, including video and audio.

This study aims to construct an audio-video generative model with minimal computational cost by leveraging pre-trained single-modal generative models for audio and video. To achieve this, we propose a novel method that guides single-modal models to cooperatively generate well-aligned samples across modalities. Specifically, given two pre-trained base diffusion models, we train a lightweight joint guidance module to adjust scores separately estimated by the base models to match the score of joint distribution over audio and video. We show that this guidance can be computed using the gradient of the optimal discriminator, which distinguishes real audio-video pairs from fake ones independently generated by the base models. Based on this analysis, we construct a joint guidance module by training this discriminator. Additionally, we adopt a loss function to stabilize the discriminator's gradient and make it work as a noise estimator, as in standard diffusion models. Empirical evaluations on several benchmark datasets demonstrate that our method improves both single-modal fidelity and multimodal alignment with relatively few parameters. The code is available at: [https://github.com/SonyResearch/MMDisCo](https://github.com/SonyResearch/MMDisCo).

We present LARP, a novel video tokenizer designed to overcome limitations in current video tokenization methods for autoregressive (AR) generative models. Unlike traditional patchwise tokenizers that directly encode local visual patches into discrete tokens, LARP introduces a holistic tokenization scheme that gathers information from the visual content using a set of learned holistic queries. This design allows LARP to capture more global and semantic representations, rather than being limited to local patch-level information. Furthermore, it offers flexibility by supporting an arbitrary number of discrete tokens, enabling adaptive and efficient tokenization based on the specific requirements of the task. To align the discrete token space with downstream AR generation tasks, LARP integrates a lightweight AR transformer as a training-time prior model that predicts the next token on its discrete latent space. By incorporating the prior model during training, LARP learns a latent space that is not only optimized for video reconstruction but is also structured in a way that is more conducive to autoregressive generation. Moreover, this process defines a sequential order for the discrete tokens, progressively pushing them toward an optimal configuration during training, ensuring smoother and more accurate AR generation at inference time. Comprehensive experiments demonstrate LARPs strong performance, achieving state-of-the-art FVD on the UCF101 class-conditional video generation benchmark. LARP enhances the compatibility of AR models with videos and opens up the potential to build unified high-fidelity multimodal large language models (MLLMs). Project page: https://hywang66.github.io/larp/

Audio-visual speech recognition (AVSR) incorporates auditory and visual modalities to improve recognition accuracy, particularly in noisy environments where audio-only speech systems are insufficient. While previous research has largely addressed audio disruptions, few studies have dealt with visual corruptions, e.g., lip occlusions or blurred videos, which are also detrimental. To address this real-world challenge, we propose CAV2vec, a novel self-supervised speech representation learning framework particularly designed to handle audio-visual joint corruption. CAV2vec employs a self-distillation approach with a corrupted prediction task, where the student model learns to predict clean targets, generated by the teacher model, with corrupted input frames. Specifically, we suggest a unimodal multi-task learning, which distills cross-modal knowledge and aligns the corrupted modalities, by predicting clean audio targets with corrupted videos, and clean video targets with corrupted audios. This strategy mitigates the dispersion in the representation space caused by corrupted modalities, leading to more reliable and robust audio-visual fusion. Our experiments on robust AVSR benchmarks demonstrate that the corrupted representation learning method significantly enhances recognition accuracy across generalized environments involving various types of corruption. Our code is available at https://github.com/sungnyun/cav2vec.

Video Multimodal Large Language Models~(MLLMs) have shown remarkable capability of understanding the video semantics on various downstream tasks. Despite the advancements, there is still a lack of systematic research on visual context representation, which refers to the scheme to select frames from a video and further select the tokens from a frame. In this paper, we explore the design space for visual context representation, and aim to improve the performance of video MLLMs by finding more effective representation schemes. Firstly, we formulate the task of visual context representation as a constrained optimization problem, and model the language modeling loss as a function of the number of frames and the number of embeddings (or tokens) per frame, given the maximum visual context window size. Then, we explore the scaling effects in frame selection and token selection respectively, and fit the corresponding function curve by conducting extensive empirical experiments. We examine the effectiveness of typical selection strategies and present empirical findings to determine the two factors. Furthermore, we study the joint effect of frame selection and token selection, and derive the optimal formula for determining the two factors. We demonstrate that the derived optimal settings show alignment with the best-performed results of empirical experiments. The data and code are available at: https://github.com/RUCAIBox/Opt-Visor.

The video-conditioned policy takes prompt videos of the desired tasks as a condition and is regarded for its prospective generalizability. Despite its promise, training a video-conditioned policy is non-trivial due to the need for abundant demonstrations. In some tasks, the expert rollouts are merely available as videos, and costly and time-consuming efforts are required to annotate action labels. To address this, we explore training video-conditioned policy on a mixture of demonstrations and unlabeled expert videos to reduce reliance on extensive manual annotation. We introduce the Joint Embedding Predictive Transformer (JEPT) to learn a video-conditioned policy through sequence modeling. JEPT is designed to jointly learn visual transition prediction and inverse dynamics. The visual transition is captured from both demonstrations and expert videos, on the basis of which the inverse dynamics learned from demonstrations is generalizable to the tasks without action labels. Experiments on a series of simulated visual control tasks evaluate that JEPT can effectively leverage the mixture dataset to learn a generalizable policy. JEPT outperforms baselines in the tasks without action-labeled data and unseen tasks. We also experimentally reveal the potential of JEPT as a simple visual priors injection approach to enhance the video-conditioned policy.

Aligning vision and language concepts at a finer level remains an essential topic of multimodal large language models (MLLMs), particularly for tasks such as referring and grounding. Existing methods, such as *proxy encoding* and *geometry encoding* genres, incorporate additional syntax to encode spatial information, imposing extra burdens when communicating between language with vision modules. In this study, we propose ClawMachine, offering a new methodology that explicitly notates each entity using **token collectives**—groups of visual tokens that collaboratively represent higher-level semantics. A hybrid perception mechanism is also explored to perceive and understand scenes from both discrete and continuous spaces. Our method unifies the prompt and answer of visual referential tasks without using additional syntax. By leveraging a joint vision-language vocabulary, ClawMachine integrates referring and grounding in an auto-regressive manner, demonstrating great potential with scaled up pre-training data. Experiments show that ClawMachine achieves superior performance on scene-level and referential understanding tasks with higher efficiency. It also exhibits the potential to integrate multi-source information for complex visual reasoning, which is beyond the capability of many MLLMs. Our code is available at https://github.com/martian422/ClawMachine.

We introduce Centaurus, a class of networks composed of generalized state-space model (SSM) blocks, where the SSM operations can be treated as tensor contractions during training. The optimal order of tensor contractions can then be systematically determined for every SSM block to maximize training efficiency. This allows more flexibility in designing SSM blocks beyond the depthwise-separable configuration commonly implemented. The new design choices will take inspiration from classical convolutional blocks including group convolutions, full convolutions, and bottleneck blocks. We architect the Centaurus network with a mixture of these blocks, to balance between network size and performance, as well as memory and computational efficiency during both training and inference. We show that this heterogeneous network design outperforms its homogeneous counterparts in raw audio processing tasks including keyword spotting, speech denoising, and automatic speech recognition (ASR). For ASR, Centaurus is the first network with competitive performance that can be made fully state-space based, without using any nonlinear recurrence (LSTMs), explicit convolutions (CNNs), or (surrogate) attention mechanism.

Large Language Models (LLMs) have allowed recent LLM-based approaches to achieve excellent performance on long-video understanding benchmarks. We investigate how extensive world knowledge and strong reasoning skills of underlying LLMs influence this strong performance. Surprisingly, we discover that LLM-based approaches can yield surprisingly good accuracy on long-video tasks with limited video information, sometimes even with no video-specific information. Building on this, we explore injecting video-specific information into an LLM-based framework. We utilize off-the-shelf vision tools to extract three object-centric information modalities from videos, and then leverage natural language as a medium for fusing this information. Our resulting Multimodal Video Understanding (MVU) framework demonstrates state-of-the-art performance across multiple video understanding benchmarks. Strong performance also on robotics domain tasks establishes its strong generality. Code: github.com/kahnchana/mvu

Visual reprogramming (VR) leverages well-developed pre-trained models (e.g., a pre-trained classifier on ImageNet) to tackle target tasks (e.g., a traffic sign recognition task), without the need for training from scratch. Despite the effectiveness of previous VR methods, all of them did not consider the adversarial robustness of reprogrammed models against adversarial attacks, which could lead to unpredictable problems in safety-crucial target tasks. In this paper, we empirically find that reprogramming pre-trained models with adversarial robustness and incorporating adversarial samples from the target task during reprogramming can both improve the adversarial robustness of reprogrammed models. Furthermore, we propose a theoretically guaranteed adversarial robustness risk upper bound for VR, which validates our empirical findings and could provide a theoretical foundation for future research. Extensive experiments demonstrate that by adopting the strategies revealed in our empirical findings, the adversarial robustness of reprogrammed models can be enhanced.