We tackle open-vocabulary 3D scene understanding by introducing a novel data generation pipeline and training framework. Our method addresses three critical requirements for effective training: precise 3D region segmentation, comprehensive textual descriptions, and sufficient dataset scale. By leveraging state-of-the-art open-vocabulary image segmentation models and region-aware Vision-Language Models, we develop an automatic pipeline that generates high-quality 3D mask-text pairs. Applying this pipeline to multiple 3D scene datasets, we create Mosaic3D-5.6M, a dataset of over 30K annotated scenes with 5.6M mask-text pairs, significantly larger than existing datasets. Building upon this data, we propose Mosaic3D, a foundation model combining a 3D encoder trained with contrastive learning and a lightweight mask decoder for open-vocabulary 3D semantic and instance segmentation. Our approach achieves state-of-the-art results on open-vocabulary 3D semantic and instance segmentation tasks including ScanNet200, Matterport3D, and ScanNet++, with ablation studies validating the effectiveness of our large-scale training data.
We introduce AutoGraph, a novel autoregressive framework for generating large attributed graphs using decoder-only transformers. At the core of our approach is a reversible "flattening" process that transforms graphs into random sequences. By sampling and learning from these sequences, AutoGraph enables transformers to model and generate complex graph structures in a manner akin to natural language. In contrast to diffusion models that rely on computationally intensive node features, our approach operates exclusively on these sequences. The sampling complexity and sequence length scale linearly with the number of edges, making AutoGraph highly scalable for generating large sparse graphs. Empirically, AutoGraph achieves state-of-the-art performance across diverse synthetic and molecular graph generation benchmarks, while delivering a 100-fold generation and a 3-fold training speedup compared to leading diffusion models. Additionally, it demonstrates promising transfer capabilities and supports substructure-conditioned generation without additional fine-tuning. By extending language modeling techniques to graph generation, this work paves the way for developing graph foundation models.
Recent advancements in foundation models have transformed computer vision, driving significant performance improvements across diverse domains, including digital histopathology. However, the advantages of domain-specific histopathology foundation models over general-purpose models for specialized tasks such as cell analysis remain underexplored. This study investigates the representation learning gap between these two categories by analyzing multi-level patch embeddings applied to cell instance segmentation and classification. We implement an encoder-decoder architecture with a consistent decoder and various encoders. These include convolutional, vision transformer (ViT), and hybrid encoders pre-trained on ImageNet-22K or LVD-142M, representing general-purpose foundation models. These are compared against ViT encoders from the recently released UNI, Virchow2, and Prov-GigaPath foundation models, trained on patches extracted from hundreds of thousands of histopathology whole-slide images. The decoder integrates patch embeddings from different encoder depths via skip connections to generate semantic and distance maps. These maps are then post-processed to create instance segmentation masks where each label corresponds to an individual cell and to perform cell-type classification. All encoders remain frozen during training to assess their pre-trained feature extraction capabilities. Using the PanNuke and CoNIC histopathology datasets, and the newly introduced Nissl-stained CytoDArk0 dataset for brain cytoarchitecture studies, we evaluate instance-level detection, segmentation accuracy, and cell-type classification. This study provides insights into the comparative strengths and limitations of general-purpose vs. histopathology foundation models, offering guidance for model selection in cell-focused histopathology and brain cytoarchitecture analysis workflows.
X-ray observing facilities, such as the Chandra X-ray Observatory and the eROSITA, have detected millions of astronomical sources associated with high-energy phenomena. The arrival of photons as a function of time follows a Poisson process and can vary by orders-of-magnitude, presenting obstacles for common tasks such as source classification, physical property derivation, and anomaly detection. Previous work has either failed to directly capture the Poisson nature of the data or only focuses on Poisson rate function reconstruction. In this work, we present Poisson Process AutoDecoder (PPAD). PPAD is a neural field decoder that maps fixed-length latent features to continuous Poisson rate functions across energy band and time via unsupervised learning. PPAD reconstructs the rate function and yields a representation at the same time. We demonstrate the efficacy of PPAD via reconstruction, regression, classification and anomaly detection experiments using the Chandra Source Catalog.
This paper presents a teleoperation system for controlling a redundant degree of freedom robot manipulator using human arm gestures. We propose a GRU-based Variational Autoencoder to learn a latent representation of the manipulator's configuration space, capturing its complex joint kinematics. A fully connected neural network maps human arm configurations into this latent space, allowing the system to mimic and generate corresponding manipulator trajectories in real time through the VAE decoder. The proposed method shows promising results in teleoperating the manipulator, enabling the generation of novel manipulator configurations from human features that were not present during training.
Audio-Visual Speech Recognition (AVSR) combines lip-based video with audio and can improve performance in noise, but most methods are trained only on English data. One limitation is the lack of large-scale multilingual video data, which makes it hard to train models from scratch. In this work, we propose mWhisper-Flamingo for multilingual AVSR which combines the strengths of a pre-trained audio model (Whisper) and video model (AV-HuBERT). To enable better multi-modal integration and improve the noisy multilingual performance, we introduce decoder modality dropout where the model is trained both on paired audio-visual inputs and separate audio/visual inputs. mWhisper-Flamingo achieves state-of-the-art WER on MuAViC, an AVSR dataset of 9 languages. Audio-visual mWhisper-Flamingo consistently outperforms audio-only Whisper on all languages in noisy conditions.
Large multimodal models (LMMs) demonstrate impressive capabilities in understanding images, videos, and audio beyond text. However, efficiently serving LMMs in production environments poses significant challenges due to their complex architectures and heterogeneous characteristics across their multi-stage inference pipelines. We present the first comprehensive systems analysis of two prominent LMM architectures, decoder-only and cross-attention, across six representative open-source models, revealing key systems design implications. We also present an in-depth analysis of production LMM inference traces, uncovering unique workload characteristics, including variable, heavy-tailed request distributions and bursty traffic patterns. Based on these insights, we propose ModServe, a modular LMM serving system that decouples stages for independent optimization and adaptive scaling. ModServe dynamically reconfigures stages and handles bursty traffic with modality-aware scheduling and autoscaling to meet tail latency SLOs while minimizing costs. ModServe achieves 3.3-5.5x higher throughput (leading to 25-41.3% cost saving) while meeting SLOs on a 128-GPU cluster with production traces.
This paper introduces ViNet-S, a 36MB model based on the ViNet architecture with a U-Net design, featuring a lightweight decoder that significantly reduces model size and parameters without compromising performance. Additionally, ViNet-A (148MB) incorporates spatio-temporal action localization (STAL) features, differing from traditional video saliency models that use action classification backbones. Our studies show that an ensemble of ViNet-S and ViNet-A, by averaging predicted saliency maps, achieves state-of-the-art performance on three visual-only and six audio-visual saliency datasets, outperforming transformer-based models in both parameter efficiency and real-time performance, with ViNet-S reaching over 1000fps.
Chain-of-Thought (CoT) reasoning has become a powerful framework for improving complex problem-solving capabilities in Multimodal Large Language Models (MLLMs). However, the verbose nature of textual reasoning introduces significant inefficiencies. In this work, we propose $\textbf{Heima}$ (as hidden llama), an efficient reasoning framework that leverages reasoning CoTs at hidden latent space. We design the Heima Encoder to condense each intermediate CoT into a compact, higher-level hidden representation using a single thinking token, effectively minimizing verbosity and reducing the overall number of tokens required during the reasoning process. Meanwhile, we design corresponding Heima Decoder with traditional Large Language Models (LLMs) to adaptively interpret the hidden representations into variable-length textual sequence, reconstructing reasoning processes that closely resemble the original CoTs. Experimental results across diverse reasoning MLLM benchmarks demonstrate that Heima model achieves higher generation efficiency while maintaining or even better zero-shot task accuracy. Moreover, the effective reconstruction of multimodal reasoning processes with Heima Decoder validates both the robustness and interpretability of our approach.
Multimodal Large Language Models (MLLMs) are typically based on decoder-only or cross-attention architectures. While decoder-only MLLMs outperform their cross-attention counterparts, they require significantly higher computational resources due to extensive self-attention and FFN operations on visual tokens. This raises the question: can we eliminate these expensive operations while maintaining the performance? To this end, we present a novel analysis framework to investigate the necessity of these costly operations in decoder-only MLLMs. Our framework introduces two key innovations: (1) Hollow Attention, which limits visual token interactions to local attention while maintaining visual-text associations, and (2) Probe-Activated Dynamic FFN, which selectively activates FFN parameters for visual tokens. Both methods do not require fine-tuning, which significantly enhances analysis efficiency. To assess the impact of applying these reductions across different proportions of layers, we developed a greedy search method that significantly narrows the search space. Experiments on state-of-the-art MLLMs reveal that applying our reductions to approximately half of the layers not only maintains but sometimes improves model performance, indicating significant computational redundancy in current architectures. Additionally, our method is orthogonal to existing token compression techniques, allowing for further combination to achieve greater computational reduction. Our findings may provide valuable insights for the design of more efficient future MLLMs. Our code will be publicly available at https://github.com/L-Hugh/Beyond-Token-Compression.
Designing the physical encoder is crucial for accurate image reconstruction in computational imaging (CI) systems. Currently, these systems are designed via end-to-end (E2E) optimization, where the encoder is modeled as a neural network layer and is jointly optimized with the decoder. However, the performance of E2E optimization is significantly reduced by the physical constraints imposed on the encoder. Also, since the E2E learns the parameters of the encoder by backpropagating the reconstruction error, it does not promote optimal intermediate outputs and suffers from gradient vanishing. To address these limitations, we reinterpret the concept of knowledge distillation (KD) for designing a physically constrained CI system by transferring the knowledge of a pretrained, less-constrained CI system. Our approach involves three steps: (1) Given the original CI system (student), a teacher system is created by relaxing the constraints on the student's encoder. (2) The teacher is optimized to solve a less-constrained version of the student's problem. (3) The teacher guides the training of the student through two proposed knowledge transfer functions, targeting both the encoder and the decoder feature space. The proposed method can be employed to any imaging modality since the relaxation scheme and the loss functions can be adapted according to the physical acquisition and the employed decoder. This approach was validated on three representative CI modalities: magnetic resonance, single-pixel, and compressive spectral imaging. Simulations show that a teacher system with an encoder that has a structure similar to that of the student encoder provides effective guidance. Our approach achieves significantly improved reconstruction performance and encoder design, outperforming both E2E optimization and traditional non-data-driven encoder designs.
This study presents a novel approach to human keypoint detection in low-resolution thermal images using transfer learning techniques. We introduce the first application of the Timed Up and Go (TUG) test in thermal image computer vision, establishing a new paradigm for mobility assessment. Our method leverages a MobileNetV3-Small encoder and a ViTPose decoder, trained using a composite loss function that balances latent representation alignment and heatmap accuracy. The model was evaluated using the Object Keypoint Similarity (OKS) metric from the COCO Keypoint Detection Challenge. The proposed model achieves better performance with AP, AP50, and AP75 scores of 0.861, 0.942, and 0.887 respectively, outperforming traditional supervised learning approaches like Mask R-CNN and ViTPose-Base. Moreover, our model demonstrates superior computational efficiency in terms of parameter count and FLOPS. This research lays a solid foundation for future clinical applications of thermal imaging in mobility assessment and rehabilitation monitoring.
Variational Autoencoders (VAEs) are essential tools in generative modeling and image reconstruction, with their performance heavily influenced by the encoder-decoder architecture. This study aims to improve the quality of reconstructed images by enhancing their resolution and preserving finer details, particularly when working with low-resolution inputs (16x16 pixels), where traditional VAEs often yield blurred or in-accurate results. To address this, we propose a hybrid model that combines quantum computing techniques in the VAE encoder with convolutional neural networks (CNNs) in the decoder. By upscaling the resolution from 16x16 to 32x32 during the encoding process, our approach evaluates how the model reconstructs images with enhanced resolution while maintaining key features and structures. This method tests the model's robustness in handling image reconstruction and its ability to preserve essential details despite training on lower-resolution data. We evaluate our proposed down sampling filter for Quantum VAE (Q-VAE) on the MNIST and USPS datasets and compare it with classical VAEs and a variant called Classical Direct Passing VAE (CDP-VAE), which uses windowing pooling filters in the encoding process. Performance is assessed using metrics such as the Frechet Inception Distance (FID) and Mean Squared Error (MSE), which measure the fidelity of reconstructed images. Our results demonstrate that the Q-VAE consistently outperforms both the Classical VAE and CDP-VAE, achieving significantly lower FID and MSE scores. Additionally, CDP-VAE yields better performance than C-VAE. These findings highlight the potential of quantum-enhanced VAEs to improve image reconstruction quality by enhancing resolution and preserving essential features, offering a promising direction for future applications in computer vision and synthetic data generation.
We present a novel approach centered on the decoding stage of Automatic Speech Recognition (ASR) that enhances multilingual performance, especially for low-resource languages. It utilizes a cross-lingual embedding clustering method to construct a hierarchical Softmax (H-Softmax) decoder, which enables similar tokens across different languages to share similar decoder representations. It addresses the limitations of the previous Huffman-based H-Softmax method, which relied on shallow features in token similarity assessments. Through experiments on a downsampled dataset of 15 languages, we demonstrate the effectiveness of our approach in improving low-resource multilingual ASR accuracy.
High-cardinality categorical features are a common characteristic of mixed-type tabular datasets. Existing generative model architectures struggle to learn the complexities of such data at scale, primarily due to the difficulty of parameterizing the categorical features. In this paper, we present a general variational autoencoder model, CardiCat, that can accurately fit imbalanced high-cardinality and heterogeneous tabular data. Our method substitutes one-hot encoding with regularized dual encoder-decoder embedding layers, which are jointly learned. This approach enables us to use embeddings that depend also on the other covariates, leading to a compact and homogenized parameterization of categorical features. Our model employs a considerably smaller trainable parameter space than competing methods, enabling learning at a large scale. CardiCat generates high-quality synthetic data that better represent high-cardinality and imbalanced features compared to competing VAE models for multiple real and simulated datasets.
Despite remarkable advancements in recent voice conversion (VC) systems, enhancing speaker similarity in zero-shot scenarios remains challenging. This challenge arises from the difficulty of generalizing and adapting speaker characteristics in speech within zero-shot environments, which is further complicated by mismatch between the training and inference processes. To address these challenges, we propose VoicePrompter, a robust zero-shot VC model that leverages in-context learning with voice prompts. VoicePrompter is composed of (1) a factorization method that disentangles speech components and (2) a DiT-based conditional flow matching (CFM) decoder that conditions on these factorized features and voice prompts. Additionally, (3) latent mixup is used to enhance in-context learning by combining various speaker features. This approach improves speaker similarity and naturalness in zero-shot VC by applying mixup to latent representations. Experimental results demonstrate that VoicePrompter outperforms existing zero-shot VC systems in terms of speaker similarity, speech intelligibility, and audio quality. Our demo is available at \url{https://hayeong0.github.io/VoicePrompter-demo/}.
Understanding the traversability of terrain is essential for autonomous robot navigation, particularly in unstructured environments such as natural landscapes. Although traditional methods, such as occupancy mapping, provide a basic framework, they often fail to account for the complex mobility capabilities of some platforms such as legged robots. In this work, we propose a method for estimating terrain traversability by learning from demonstrations of human walking. Our approach leverages dense, pixel-wise feature embeddings generated using the DINOv2 vision Transformer model, which are processed through an encoder-decoder MLP architecture to analyze terrain segments. The averaged feature vectors, extracted from the masked regions of interest, are used to train the model in a reconstruction-based framework. By minimizing reconstruction loss, the network distinguishes between familiar terrain with a low reconstruction error and unfamiliar or hazardous terrain with a higher reconstruction error. This approach facilitates the detection of anomalies, allowing a legged robot to navigate more effectively through challenging terrain. We run real-world experiments on the ANYmal legged robot both indoor and outdoor to prove our proposed method. The code is open-source, while video demonstrations can be found on our website: https://rpl-cs-ucl.github.io/STEPP
Machine Translation (MT) is undergoing a paradigm shift, with systems based on fine-tuned large language models (LLM) becoming increasingly competitive with traditional encoder-decoder models trained specifically for translation tasks. However, LLM-based systems are at a higher risk of generating hallucinations, which can severely undermine user's trust and safety. Most prior research on hallucination mitigation focuses on traditional MT models, with solutions that involve post-hoc mitigation - detecting hallucinated translations and re-translating them. While effective, this approach introduces additional complexity in deploying extra tools in production and also increases latency. To address these limitations, we propose a method that intrinsically learns to mitigate hallucinations during the model training phase. Specifically, we introduce a data creation framework to generate hallucination focused preference datasets. Fine-tuning LLMs on these preference datasets reduces the hallucination rate by an average of 96% across five language pairs, while preserving overall translation quality. In a zero-shot setting our approach reduces hallucinations by 89% on an average across three unseen target languages.
We propose a modular modeling framework designed to enhance the capture and validation of uncertainty in autonomous vehicle (AV) trajectory prediction. Departing from traditional deterministic methods, our approach employs a flexible, end-to-end differentiable probabilistic encoder-decoder architecture. This modular design allows the encoder and decoder to be trained independently, enabling seamless adaptation to diverse traffic scenarios without retraining the entire system. Our key contributions include: (1) a probabilistic heatmap predictor that generates context-aware occupancy grids for dynamic forecasting, (2) a modular training approach that supports independent component training and flexible adaptation, and (3) a structured validation scheme leveraging uncertainty metrics to evaluate robustness under high-risk conditions. To highlight the benefits of our framework, we benchmark it against an end-to-end baseline, demonstrating faster convergence, improved stability, and flexibility. Experimental results validate these advantages, showcasing the capacity of the framework to efficiently handle complex scenarios while ensuring reliable predictions and robust uncertainty representation. This modular design offers significant practical utility and scalability for real-world autonomous driving applications.
Applying Multimodal Large Language Models (MLLMs) to video understanding presents significant challenges due to the need to model temporal relations across frames. Existing approaches adopt either implicit temporal modeling, relying solely on the LLM decoder, or explicit temporal modeling, employing auxiliary temporal encoders. To investigate this debate between the two paradigms, we propose the Stackable Temporal Encoder (STE). STE enables flexible explicit temporal modeling with adjustable temporal receptive fields and token compression ratios. Using STE, we systematically compare implicit and explicit temporal modeling across dimensions such as overall performance, token compression effectiveness, and temporal-specific understanding. We also explore STE's design considerations and broader impacts as a plug-in module and in image modalities. Our findings emphasize the critical role of explicit temporal modeling, providing actionable insights to advance video MLLMs.