Abstract:The scaling up has brought tremendous success in the fields of vision and language in recent years. When it comes to audio, however, researchers encounter a major challenge in scaling up the training data, as most natural audio contains diverse interfering signals. To address this limitation, we introduce Omni-modal Sound Separation (OmniSep), a novel framework capable of isolating clean soundtracks based on omni-modal queries, encompassing both single-modal and multi-modal composed queries. Specifically, we introduce the Query-Mixup strategy, which blends query features from different modalities during training. This enables OmniSep to optimize multiple modalities concurrently, effectively bringing all modalities under a unified framework for sound separation. We further enhance this flexibility by allowing queries to influence sound separation positively or negatively, facilitating the retention or removal of specific sounds as desired. Finally, OmniSep employs a retrieval-augmented approach known as Query-Aug, which enables open-vocabulary sound separation. Experimental evaluations on MUSIC, VGGSOUND-CLEAN+, and MUSIC-CLEAN+ datasets demonstrate effectiveness of OmniSep, achieving state-of-the-art performance in text-, image-, and audio-queried sound separation tasks. For samples and further information, please visit the demo page at \url{https://omnisep.github.io/}.
Abstract:Generating music that aligns with the visual content of a video has been a challenging task, as it requires a deep understanding of visual semantics and involves generating music whose melody, rhythm, and dynamics harmonize with the visual narratives. This paper presents MuVi, a novel framework that effectively addresses these challenges to enhance the cohesion and immersive experience of audio-visual content. MuVi analyzes video content through a specially designed visual adaptor to extract contextually and temporally relevant features. These features are used to generate music that not only matches the video's mood and theme but also its rhythm and pacing. We also introduce a contrastive music-visual pre-training scheme to ensure synchronization, based on the periodicity nature of music phrases. In addition, we demonstrate that our flow-matching-based music generator has in-context learning ability, allowing us to control the style and genre of the generated music. Experimental results show that MuVi demonstrates superior performance in both audio quality and temporal synchronization. The generated music video samples are available at https://muvi-v2m.github.io.
Abstract: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 related code, demos, and pre-trained models are available at https://github.com/jishengpeng/WavTokenizer.
Abstract:Frequency-modulated continuous-wave (FMCW) radar plays a pivotal role in the field of remote sensing. The increasing degree of FMCW radar deployment has increased the mutual interference, which weakens the detection capabilities of radars and threatens reliability and safety of systems. In this paper, a novel FMCW radar interference mitigation (RIM) method, termed as RIMformer, is proposed by using an end-to-end Transformer-based structure. In the RIMformer, a dual multi-head self-attention mechanism is proposed to capture the correlations among the distinct distance elements of intermediate frequency (IF) signals. Additionally, an improved convolutional block is integrated to harness the power of convolution for extracting local features. The architecture is designed to process time-domain IF signals in an end-to-end manner, thereby avoiding the need for additional manual data processing steps. The improved decoder structure ensures the parallelization of the network to increase its computational efficiency. Simulation and measurement experiments are carried out to validate the accuracy and effectiveness of the proposed method. The results show that the proposed RIMformer can effectively mitigate interference and restore the target signals.
Abstract:Recently, human-computer interaction with various modalities has shown promising applications, like GPT-4o and Gemini. Given the foundational role of multimodal joint representation in understanding and generation pipelines, high-quality omni joint representations would be a step toward co-processing more diverse multimodal information. In this work, we present OmniBind, large-scale multimodal joint representation models ranging in scale from 7 billion to 30 billion parameters, which support 3D, audio, image, and language inputs. Due to the scarcity of data pairs across all modalities, instead of training large models from scratch, we propose remapping and binding the spaces of various pre-trained specialist models together. This approach enables "scaling up" by indirectly increasing the model parameters and the amount of seen data. To effectively integrate various spaces, we dynamically assign weights to different spaces by learning routers with two objectives: cross-modal overall alignment and language representation decoupling. Notably, since binding and routing spaces both only require lightweight networks, OmniBind is extremely training-efficient. Learning the largest 30B model requires merely unpaired unimodal data and approximately 3 days on a single 8-4090 node. Extensive experiments demonstrate the versatility and superiority of OmniBind as an omni representation model, highlighting its great potential for diverse applications, such as any-query and composable multimodal understanding.
Abstract:Unified multi-model representation spaces are the foundation of multimodal understanding and generation. However, the billions of model parameters and catastrophic forgetting problems make it challenging to further enhance pre-trained unified spaces. In this work, we propose FreeBind, an idea that treats multimodal representation spaces as basic units, and freely augments pre-trained unified space by integrating knowledge from extra expert spaces via "space bonds". Specifically, we introduce two kinds of basic space bonds: 1) Space Displacement Bond and 2) Space Combination Bond. Based on these basic bonds, we design Complex Sequential & Parallel Bonds to effectively integrate multiple spaces simultaneously. Benefiting from the modularization concept, we further propose a coarse-to-fine customized inference strategy to flexibly adjust the enhanced unified space for different purposes. Experimentally, we bind ImageBind with extra image-text and audio-text expert spaces, resulting in three main variants: ImageBind++, InternVL_IB, and InternVL_IB++. These resulting spaces outperform ImageBind on 5 audio-image-text downstream tasks across 9 datasets. Moreover, via customized inference, it even surpasses the advanced audio-text and image-text expert spaces.
Abstract:Unified multi-model representation spaces are the foundation of multimodal understanding and generation. However, the billions of model parameters and catastrophic forgetting problems make it challenging to further enhance pre-trained unified spaces. In this work, we propose Molecule-Space, an idea that treats multimodal representation spaces as "molecules", and augments pre-trained unified space by integrating knowledge from extra expert spaces via "molecules space reactions". Specifically, we introduce two kinds of basic space reactions: 1) Space Displacement Reaction and 2) Space Combination Reaction. Based on these defined basic reactions, we design Complex Sequential & Parallel Reactions to effectively integrate multiple spaces simultaneously. Benefiting from the modularization concept, we further propose a coarse-to-fine customized inference strategy to flexibly adjust the enhanced unified space for different purposes. Experimentally, we fuse the audio-image-text space of ImageBind with the image-text and audio-text expert spaces. The resulting space outperforms ImageBind on 5 downstream tasks across 9 datasets. Moreover, via customized inference, it even surpasses the used image-text and audio-text expert spaces.
Abstract:Detecting polyps through colonoscopy is an important task in medical image segmentation, which provides significant assistance and reference value for clinical surgery. However, accurate segmentation of polyps is a challenging task due to two main reasons. Firstly, polyps exhibit various shapes and colors. Secondly, the boundaries between polyps and their normal surroundings are often unclear. Additionally, significant differences between different datasets lead to limited generalization capabilities of existing methods. To address these issues, we propose a segmentation model based on Prompt-Mamba, which incorporates the latest Vision-Mamba and prompt technologies. Compared to previous models trained on the same dataset, our model not only maintains high segmentation accuracy on the validation part of the same dataset but also demonstrates superior accuracy on unseen datasets, exhibiting excellent generalization capabilities. Notably, we are the first to apply the Vision-Mamba architecture to polyp segmentation and the first to utilize prompt technology in a polyp segmentation model. Our model efficiently accomplishes segmentation tasks, surpassing previous state-of-the-art methods by an average of 5% across six datasets. Furthermore, we have developed multiple versions of our model with scaled parameter counts, achieving better performance than previous models even with fewer parameters. Our code and trained weights will be released soon.
Abstract:Large pre-trained models with their numerous model parameters and extensive training datasets have shown excellent performance in various tasks. Many publicly available medical image datasets do not have a sufficient amount of data so there are few large-scale models in medical imaging. We propose a large-scale Tumor Segmentation Foundation Model (TSFM) with 1.6 billion parameters using Resblock-backbone and Transformer-bottleneck,which has good transfer ability for downstream tasks. To make TSFM exhibit good performance in tumor segmentation, we make full use of the strong spatial correlation between tumors and organs in the medical image, innovatively fuse 7 tumor datasets and 3 multi-organ datasets to build a 3D medical dataset pool, including 2779 cases with totally 300k medical images, whose size currently exceeds many other single publicly available datasets. TSFM is the pre-trained model for medical image segmentation, which also can be transferred to multiple downstream tasks for fine-tuning learning. The average performance of our pre-trained model is 2% higher than that of nnU-Net across various tumor types. In the transfer learning task, TSFM only needs 5% training epochs of nnU-Net to achieve similar performance and can surpass nnU-Net by 2% on average with 10% training epoch. Pre-trained TSFM and its code will be released soon.
Abstract:With the rapid advancement of artificial intelligence technology, the usage of machine learning models is gradually becoming part of our daily lives. High-quality models rely not only on efficient optimization algorithms but also on the training and learning processes built upon vast amounts of data and computational power. However, in practice, due to various challenges such as limited computational resources and data privacy concerns, users in need of models often cannot train machine learning models locally. This has led them to explore alternative approaches such as outsourced learning and federated learning. While these methods address the feasibility of model training effectively, they introduce concerns about the trustworthiness of the training process since computations are not performed locally. Similarly, there are trustworthiness issues associated with outsourced model inference. These two problems can be summarized as the trustworthiness problem of model computations: How can one verify that the results computed by other participants are derived according to the specified algorithm, model, and input data? To address this challenge, verifiable machine learning (VML) has emerged. This paper presents a comprehensive survey of zero-knowledge proof-based verifiable machine learning (ZKP-VML) technology. We first analyze the potential verifiability issues that may exist in different machine learning scenarios. Subsequently, we provide a formal definition of ZKP-VML. We then conduct a detailed analysis and classification of existing works based on their technical approaches. Finally, we discuss the key challenges and future directions in the field of ZKP-based VML.