We introduce a project that revives a piece of 15th-century Korean court music, Chihwapyeong and Chwipunghyeong, composed upon the poem Songs of the Dragon Flying to Heaven. One of the earliest examples of Jeongganbo, a Korean musical notation system, the remaining version only consists of a rudimentary melody. Our research team, commissioned by the National Gugak (Korean Traditional Music) Center, aimed to transform this old melody into a performable arrangement for a six-part ensemble. Using Jeongganbo data acquired through bespoke optical music recognition, we trained a BERT-like masked language model and an encoder-decoder transformer model. We also propose an encoding scheme that strictly follows the structure of Jeongganbo and denotes note durations as positions. The resulting machine-transformed version of Chihwapyeong and Chwipunghyeong were evaluated by experts and performed by the Court Music Orchestra of National Gugak Center. Our work demonstrates that generative models can successfully be applied to traditional music with limited training data if combined with careful design.
Representing symbolic music with compound tokens, where each token consists of several different sub-tokens representing a distinct musical feature or attribute, offers the advantage of reducing sequence length. While previous research has validated the efficacy of compound tokens in music sequence modeling, predicting all sub-tokens simultaneously can lead to suboptimal results as it may not fully capture the interdependencies between them. We introduce the Nested Music Transformer (NMT), an architecture tailored for decoding compound tokens autoregressively, similar to processing flattened tokens, but with low memory usage. The NMT consists of two transformers: the main decoder that models a sequence of compound tokens and the sub-decoder for modeling sub-tokens of each compound token. The experiment results showed that applying the NMT to compound tokens can enhance the performance in terms of better perplexity in processing various symbolic music datasets and discrete audio tokens from the MAESTRO dataset.
Most advanced visual grounding methods rely on Transformers for visual-linguistic feature fusion. However, these Transformer-based approaches encounter a significant drawback: the computational costs escalate quadratically due to the self-attention mechanism in the Transformer Encoder, particularly when dealing with high-resolution images or long context sentences. This quadratic increase in computational burden restricts the applicability of visual grounding to more intricate scenes, such as conversation-based reasoning segmentation, which involves lengthy language expressions. In this paper, we propose an efficient and effective multi-task visual grounding (EEVG) framework based on Transformer Decoder to address this issue, which reduces the cost in both language and visual aspects. In the language aspect, we employ the Transformer Decoder to fuse visual and linguistic features, where linguistic features are input as memory and visual features as queries. This allows fusion to scale linearly with language expression length. In the visual aspect, we introduce a parameter-free approach to reduce computation by eliminating background visual tokens based on attention scores. We then design a light mask head to directly predict segmentation masks from the remaining sparse feature maps. Extensive results and ablation studies on benchmarks demonstrate the efficiency and effectiveness of our approach. Code is available in https://github.com/chenwei746/EEVG.
Point cloud compression is essential to experience volumetric multimedia as it drastically reduces the required streaming data rates. Point attributes, specifically colors, extend the challenge of lossy compression beyond geometric representation to achieving joint reconstruction of texture and geometry. State-of-the-art methods separate geometry and attributes to compress them individually. This comes at a computational cost, requiring an encoder and a decoder for each modality. Additionally, as attribute compression methods require the same geometry for encoding and decoding, the encoder emulates the decoder-side geometry reconstruction as an input step to project and compress the attributes. In this work, we propose to learn joint compression of geometry and attributes using a single, adaptive autoencoder model, embedding both modalities into a unified latent space which is then entropy encoded. Key to the technique is to replace the search for trade-offs between rate, attribute quality and geometry quality, through conditioning the model on the desired qualities of both modalities, bypassing the need for training model ensembles. To differentiate important point cloud regions during encoding or to allow view-dependent compression for user-centered streaming, conditioning is pointwise, which allows for local quality and rate variation. Our evaluation shows comparable performance to state-of-the-art compression methods for geometry and attributes, while reducing complexity compared to related compression methods.
Throughout rapid development of multimodal large language models, a crucial ingredient is a fair and accurate evaluation of their multimodal comprehension abilities. Although Visual Question Answering (VQA) could serve as a developed test field, limitations of VQA evaluation, like the inflexible pattern of Exact Match, have hindered MLLMs from demonstrating their real capability and discourage rich responses. Therefore, this paper proposes the use of semantics-based evaluators for assessing unconstrained open-ended responses on VQA datasets. As characteristics of VQA have made such evaluation significantly different than the traditional Semantic Textual Similarity (STS) task, to systematically analyze the behaviour and compare the performance of various evaluators including LLM-based ones, we proposes three key properties, i.e., Alignment, Consistency and Generalization, and a corresponding dataset Assessing VQA Evaluators (AVE) to facilitate analysis. In addition, this paper proposes a Semantically Flexible VQA Evaluator (SFVE) with meticulous design based on the unique features of VQA evaluation. Experimental results verify the feasibility of model-based VQA evaluation and effectiveness of the proposed evaluator that surpasses existing semantic evaluators by a large margin. The proposed training scheme generalizes to both the BERT-like encoders and decoder-only LLM.
3D object detection is essential for understanding 3D scenes. Contemporary techniques often require extensive annotated training data, yet obtaining point-wise annotations for point clouds is time-consuming and laborious. Recent developments in semi-supervised methods seek to mitigate this problem by employing a teacher-student framework to generate pseudo-labels for unlabeled point clouds. However, these pseudo-labels frequently suffer from insufficient diversity and inferior quality. To overcome these hurdles, we introduce an Agent-based Diffusion Model for Semi-supervised 3D Object Detection (Diff3DETR). Specifically, an agent-based object query generator is designed to produce object queries that effectively adapt to dynamic scenes while striking a balance between sampling locations and content embedding. Additionally, a box-aware denoising module utilizes the DDIM denoising process and the long-range attention in the transformer decoder to refine bounding boxes incrementally. Extensive experiments on ScanNet and SUN RGB-D datasets demonstrate that Diff3DETR outperforms state-of-the-art semi-supervised 3D object detection methods.
Current large language models (LLMs) primarily utilize next-token prediction method for inference, which significantly impedes their processing speed. In this paper, we introduce a novel inference methodology termed next-sentence prediction, aiming at enhancing the inference efficiency of LLMs. We present Sentence Variational Autoencoder (SentenceVAE), which includes a Sentence Encoder to compress multiple tokens in a sentence into a single token, and a Sentence Decoder to reconstruct it. By integrating SentenceVAE into the input and output layers of LLMs, we develop Sentence-level LLMs (SLLMs) that employ a sentence-by-sentence inference method. In addition, the SentenceVAE module of SLLMs can maintain the integrity of the original semantic content by segmenting the context into sentences, thereby improving accuracy while boosting inference speed. Moreover, compared to previous LLMs, SLLMs process fewer tokens over equivalent context length, significantly reducing memory demands for self-attention computation and facilitating the handling of longer context. Extensive experiments on Wanjuan dataset have revealed that the proposed method can accelerate inference speed by 204~365%, reduce perplexity (PPL) to 46~75% of its original metric, and decrease memory overhead by 86~91% for the equivalent context length, compared to previous token-by-token methods.
Blind iris images, which result from unknown degradation during the process of iris recognition at long distances, often lead to decreased iris recognition rates. Currently, little existing literature offers a solution to this problem. In response, we propose a prior embedding-driven architecture for long distance blind iris recognition. We first proposed a blind iris image restoration network called Iris-PPRGAN. To effectively restore the texture of the blind iris, Iris-PPRGAN includes a Generative Adversarial Network (GAN) used as a Prior Decoder, and a DNN used as the encoder. To extract iris features more efficiently, we then proposed a robust iris classifier by modifying the bottleneck module of InsightFace, which called Insight-Iris. A low-quality blind iris image is first restored by Iris-PPRGAN, then the restored iris image undergoes recognition via Insight-Iris. Experimental results on the public CASIA-Iris-distance dataset demonstrate that our proposed method significantly superior results to state-of-the-art blind iris restoration methods both quantitatively and qualitatively, Specifically, the recognition rate for long-distance blind iris images reaches 90% after processing with our methods, representing an improvement of approximately ten percentage points compared to images without restoration.
The increasing cost of training machine learning (ML) models has led to the inclusion of new parties to the training pipeline, such as users who contribute training data and companies that provide computing resources. This involvement of such new parties in the ML training process has introduced new attack surfaces for an adversary to exploit. A recent attack in this domain is the model hijacking attack, whereby an adversary hijacks a victim model to implement their own -- possibly malicious -- hijacking tasks. However, the scope of the model hijacking attack is so far limited to the homogeneous-modality tasks. In this paper, we transform the model hijacking attack into a more general multimodal setting, where the hijacking and original tasks are performed on data of different modalities. Specifically, we focus on the setting where an adversary implements a natural language processing (NLP) hijacking task into an image classification model. To mount the attack, we propose a novel encoder-decoder based framework, namely the Blender, which relies on advanced image and language models. Experimental results show that our modal hijacking attack achieves strong performances in different settings. For instance, our attack achieves 94%, 94%, and 95% attack success rate when using the Sogou news dataset to hijack STL10, CIFAR-10, and MNIST classifiers.
Accurate text segmentation results are crucial for text-related generative tasks, such as text image generation, text editing, text removal, and text style transfer. Recently, some scene text segmentation methods have made significant progress in segmenting regular text. However, these methods perform poorly in scenarios containing artistic text. Therefore, this paper focuses on the more challenging task of artistic text segmentation and constructs a real artistic text segmentation dataset. One challenge of the task is that the local stroke shapes of artistic text are changeable with diversity and complexity. We propose a decoder with the layer-wise momentum query to prevent the model from ignoring stroke regions of special shapes. Another challenge is the complexity of the global topological structure. We further design a skeleton-assisted head to guide the model to focus on the global structure. Additionally, to enhance the generalization performance of the text segmentation model, we propose a strategy for training data synthesis, based on the large multi-modal model and the diffusion model. Experimental results show that our proposed method and synthetic dataset can significantly enhance the performance of artistic text segmentation and achieve state-of-the-art results on other public datasets.
Recent approaches have shown that large-scale vision-language models such as CLIP can improve semantic segmentation performance. These methods typically aim for pixel-level vision-language alignment, but often rely on low resolution image features from CLIP, resulting in class ambiguities along boundaries. Moreover, the global scene representations in CLIP text embeddings do not directly correlate with the local and detailed pixel-level features, making meaningful alignment more difficult. To address these limitations, we introduce MTA-CLIP, a novel framework employing mask-level vision-language alignment. Specifically, we first propose Mask-Text Decoder that enhances the mask representations using rich textual data with the CLIP language model. Subsequently, it aligns mask representations with text embeddings using Mask-to-Text Contrastive Learning. Furthermore, we introduce MaskText Prompt Learning, utilizing multiple context-specific prompts for text embeddings to capture diverse class representations across masks. Overall, MTA-CLIP achieves state-of-the-art, surpassing prior works by an average of 2.8% and 1.3% on on standard benchmark datasets, ADE20k and Cityscapes, respectively.
Medical image segmentation involves identifying and separating object instances in a medical image to delineate various tissues and structures, a task complicated by the significant variations in size, shape, and density of these features. Convolutional neural networks (CNNs) have traditionally been used for this task but have limitations in capturing long-range dependencies. Transformers, equipped with self-attention mechanisms, aim to address this problem. However, in medical image segmentation it is beneficial to merge both local and global features to effectively integrate feature maps across various scales, capturing both detailed features and broader semantic elements for dealing with variations in structures. In this paper, we introduce MSA$^2$Net, a new deep segmentation framework featuring an expedient design of skip-connections. These connections facilitate feature fusion by dynamically weighting and combining coarse-grained encoder features with fine-grained decoder feature maps. Specifically, we propose a Multi-Scale Adaptive Spatial Attention Gate (MASAG), which dynamically adjusts the receptive field (Local and Global contextual information) to ensure that spatially relevant features are selectively highlighted while minimizing background distractions. Extensive evaluations involving dermatology, and radiological datasets demonstrate that our MSA$^2$Net outperforms state-of-the-art (SOTA) works or matches their performance. The source code is publicly available at https://github.com/xmindflow/MSA-2Net.
Potato yield is an important metric for farmers to further optimize their cultivation practices. Potato yield can be estimated on a harvester using an RGB-D camera that can estimate the three-dimensional (3D) volume of individual potato tubers. A challenge, however, is that the 3D shape derived from RGB-D images is only partially completed, underestimating the actual volume. To address this issue, we developed a 3D shape completion network, called CoRe++, which can complete the 3D shape from RGB-D images. CoRe++ is a deep learning network that consists of a convolutional encoder and a decoder. The encoder compresses RGB-D images into latent vectors that are used by the decoder to complete the 3D shape using the deep signed distance field network (DeepSDF). To evaluate our CoRe++ network, we collected partial and complete 3D point clouds of 339 potato tubers on an operational harvester in Japan. On the 1425 RGB-D images in the test set (representing 51 unique potato tubers), our network achieved a completion accuracy of 2.8 mm on average. For volumetric estimation, the root mean squared error (RMSE) was 22.6 ml, and this was better than the RMSE of the linear regression (31.1 ml) and the base model (36.9 ml). We found that the RMSE can be further reduced to 18.2 ml when performing the 3D shape completion in the center of the RGB-D image. With an average 3D shape completion time of 10 milliseconds per tuber, we can conclude that CoRe++ is both fast and accurate enough to be implemented on an operational harvester for high-throughput potato yield estimation. CoRe++'s high-throughput and accurate processing allows it to be applied to other tuber, fruit and vegetable crops, thereby enabling versatile, accurate and real-time yield monitoring in precision agriculture. Our code, network weights and dataset are publicly available at https://github.com/UTokyo-FieldPhenomics-Lab/corepp.git.
Multimodal machine learning models that combine visual and textual data are increasingly being deployed in critical applications, raising significant safety and security concerns due to their vulnerability to adversarial attacks. This paper presents an effective strategy to enhance the robustness of multimodal image captioning models against such attacks. By leveraging the Fast Gradient Sign Method (FGSM) to generate adversarial examples and incorporating adversarial training techniques, we demonstrate improved model robustness on two benchmark datasets: Flickr8k and COCO. Our findings indicate that selectively training only the text decoder of the multimodal architecture shows performance comparable to full adversarial training while offering increased computational efficiency. This targeted approach suggests a balance between robustness and training costs, facilitating the ethical deployment of multimodal AI systems across various domains.
Environment prediction frameworks are critical for the safe navigation of autonomous vehicles (AVs) in dynamic settings. LiDAR-generated occupancy grid maps (L-OGMs) offer a robust bird's-eye view for the scene representation, enabling self-supervised joint scene predictions while exhibiting resilience to partial observability and perception detection failures. Prior approaches have focused on deterministic L-OGM prediction architectures within the grid cell space. While these methods have seen some success, they frequently produce unrealistic predictions and fail to capture the stochastic nature of the environment. Additionally, they do not effectively integrate additional sensor modalities present in AVs. Our proposed framework, Latent Occupancy Prediction (LOPR), performs stochastic L-OGM prediction in the latent space of a generative architecture and allows for conditioning on RGB cameras, maps, and planned trajectories. We decode predictions using either a single-step decoder, which provides high-quality predictions in real-time, or a diffusion-based batch decoder, which can further refine the decoded frames to address temporal consistency issues and reduce compression losses. Our experiments on the nuScenes and Waymo Open datasets show that all variants of our approach qualitatively and quantitatively outperform prior approaches.
The rapid development of neural text-to-speech (TTS) systems enabled its usage in other areas of natural language processing such as automatic speech recognition (ASR) or spoken language translation (SLT). Due to the large number of different TTS architectures and their extensions, selecting which TTS systems to use for synthetic data creation is not an easy task. We use the comparison of five different TTS decoder architectures in the scope of synthetic data generation to show the impact on CTC-based speech recognition training. We compare the recognition results to computable metrics like NISQA MOS and intelligibility, finding that there are no clear relations to the ASR performance. We also observe that for data generation auto-regressive decoding performs better than non-autoregressive decoding, and propose an approach to quantify TTS generalization capabilities.
The segmentation of ultra-high resolution images poses challenges such as loss of spatial information or computational inefficiency. In this work, a novel approach that combines encoder-decoder architectures with domain decomposition strategies to address these challenges is proposed. Specifically, a domain decomposition-based U-Net (DDU-Net) architecture is introduced, which partitions input images into non-overlapping patches that can be processed independently on separate devices. A communication network is added to facilitate inter-patch information exchange to enhance the understanding of spatial context. Experimental validation is performed on a synthetic dataset that is designed to measure the effectiveness of the communication network. Then, the performance is tested on the DeepGlobe land cover classification dataset as a real-world benchmark data set. The results demonstrate that the approach, which includes inter-patch communication for images divided into $16\times16$ non-overlapping subimages, achieves a $2-3\,\%$ higher intersection over union (IoU) score compared to the same network without inter-patch communication. The performance of the network which includes communication is equivalent to that of a baseline U-Net trained on the full image, showing that our model provides an effective solution for segmenting ultra-high-resolution images while preserving spatial context. The code is available at https://github.com/corne00/DDU-Net.
Large-scale pre-training methodologies for chemical language models represent a breakthrough in cheminformatics. These methods excel in tasks such as property prediction and molecule generation by learning contextualized representations of input tokens through self-supervised learning on large unlabeled corpora. Typically, this involves pre-training on unlabeled data followed by fine-tuning on specific tasks, reducing dependence on annotated datasets and broadening chemical language representation understanding. This paper introduces a large encoder-decoder chemical foundation models pre-trained on a curated dataset of 91 million SMILES samples sourced from PubChem, which is equivalent to 4 billion of molecular tokens. The proposed foundation model supports different complex tasks, including quantum property prediction, and offer flexibility with two main variants (289M and $8\times289M$). Our experiments across multiple benchmark datasets validate the capacity of the proposed model in providing state-of-the-art results for different tasks. We also provide a preliminary assessment of the compositionality of the embedding space as a prerequisite for the reasoning tasks. We demonstrate that the produced latent space is separable compared to the state-of-the-art with few-shot learning capabilities.
Point tracking is a challenging task in computer vision, aiming to establish point-wise correspondence across long video sequences. Recent advancements have primarily focused on temporal modeling techniques to improve local feature similarity, often overlooking the valuable semantic consistency inherent in tracked points. In this paper, we introduce a novel approach leveraging language embeddings to enhance the coherence of frame-wise visual features related to the same object. Our proposed method, termed autogenic language embedding for visual feature enhancement, strengthens point correspondence in long-term sequences. Unlike existing visual-language schemes, our approach learns text embeddings from visual features through a dedicated mapping network, enabling seamless adaptation to various tracking tasks without explicit text annotations. Additionally, we introduce a consistency decoder that efficiently integrates text tokens into visual features with minimal computational overhead. Through enhanced visual consistency, our approach significantly improves tracking trajectories in lengthy videos with substantial appearance variations. Extensive experiments on widely-used tracking benchmarks demonstrate the superior performance of our method, showcasing notable enhancements compared to trackers relying solely on visual cues.
Deep learning has revolutionized medical image segmentation, but it relies heavily on high-quality annotations. The time, cost and expertise required to label images at the pixel-level for each new task has slowed down widespread adoption of the paradigm. We propose Pix2Rep, a self-supervised learning (SSL) approach for few-shot segmentation, that reduces the manual annotation burden by learning powerful pixel-level representations directly from unlabeled images. Pix2Rep is a novel pixel-level loss and pre-training paradigm for contrastive SSL on whole images. It is applied to generic encoder-decoder deep learning backbones (e.g., U-Net). Whereas most SSL methods enforce invariance of the learned image-level representations under intensity and spatial image augmentations, Pix2Rep enforces equivariance of the pixel-level representations. We demonstrate the framework on a task of cardiac MRI segmentation. Results show improved performance compared to existing semi- and self-supervised approaches; and a 5-fold reduction in the annotation burden for equivalent performance versus a fully supervised U-Net baseline. This includes a 30% (resp. 31%) DICE improvement for one-shot segmentation under linear-probing (resp. fine-tuning). Finally, we also integrate the novel Pix2Rep concept with the Barlow Twins non-contrastive SSL, which leads to even better segmentation performance.