Recent proposals for quantum generative adversarial networks (GANs) suffer from the issue of mode collapse, analogous to classical GANs, wherein the distribution learnt by the GAN fails to capture the high mode complexities of the target distribution. Mode collapse can arise due to the use of uninformed prior distributions in the generative learning task. To alleviate the issue of mode collapse for quantum GANs, this work presents a novel \textbf{hybrid quantum-classical generative model}, the VAE-QWGAN, which combines the strengths of a classical Variational AutoEncoder (VAE) with a hybrid Quantum Wasserstein GAN (QWGAN). The VAE-QWGAN fuses the VAE decoder and QWGAN generator into a single quantum model, and utilizes the VAE encoder for data-dependant latent vector sampling during training. This in turn, enhances the diversity and quality of generated images. To generate new data from the trained model at inference, we sample from a Gaussian mixture model (GMM) prior that is learnt on the latent vectors generated during training. We conduct extensive experiments for image generation QGANs on MNIST/Fashion-MNIST datasets and compute a range of metrics that measure the diversity and quality of generated samples. We show that VAE-QWGAN demonstrates significant improvement over existing QGAN approaches.
Recently, biological perception has been a powerful tool for handling the camouflaged object detection (COD) task. However, most existing methods are heavily dependent on the local spatial information of diverse scales from convolutional operations to optimize initial features. A commonly neglected point in these methods is the long-range dependencies between feature pixels from different scale spaces that can help the model build a global structure of the object, inducing a more precise image representation. In this paper, we propose a novel Global-Local Collaborative Optimization Network, called GLCONet. Technically, we first design a collaborative optimization strategy from the perspective of multi-source perception to simultaneously model the local details and global long-range relationships, which can provide features with abundant discriminative information to boost the accuracy in detecting camouflaged objects. Furthermore, we introduce an adjacent reverse decoder that contains cross-layer aggregation and reverse optimization to integrate complementary information from different levels for generating high-quality representations. Extensive experiments demonstrate that the proposed GLCONet method with different backbones can effectively activate potentially significant pixels in an image, outperforming twenty state-of-the-art methods on three public COD datasets. The source code is available at: \https://github.com/CSYSI/GLCONet.
Robots interacting with humans through natural language can unlock numerous applications such as Referring Grasp Synthesis (RGS). Given a text query, RGS determines a stable grasp pose to manipulate the referred object in the robot's workspace. RGS comprises two steps: visual grounding and grasp pose estimation. Recent studies leverage powerful Vision-Language Models (VLMs) for visually grounding free-flowing natural language in real-world robotic execution. However, comparisons in complex, cluttered environments with multiple instances of the same object are lacking. This paper introduces HiFi-CS, featuring hierarchical application of Featurewise Linear Modulation (FiLM) to fuse image and text embeddings, enhancing visual grounding for complex attribute rich text queries encountered in robotic grasping. Visual grounding associates an object in 2D/3D space with natural language input and is studied in two scenarios: Closed and Open Vocabulary. HiFi-CS features a lightweight decoder combined with a frozen VLM and outperforms competitive baselines in closed vocabulary settings while being 100x smaller in size. Our model can effectively guide open-set object detectors like GroundedSAM to enhance open-vocabulary performance. We validate our approach through real-world RGS experiments using a 7-DOF robotic arm, achieving 90.33\% visual grounding accuracy in 15 tabletop scenes. Our codebase is provided here: https://github.com/vineet2104/hifics
In extreme scenarios such as nighttime or low-visibility environments, achieving reliable perception is critical for applications like autonomous driving, robotics, and surveillance. Multi-modality image fusion, particularly integrating infrared imaging, offers a robust solution by combining complementary information from different modalities to enhance scene understanding and decision-making. However, current methods face significant limitations: GAN-based approaches often produce blurry images that lack fine-grained details, while AE-based methods may introduce bias toward specific modalities, leading to unnatural fusion results. To address these challenges, we propose DAE-Fuse, a novel two-phase discriminative autoencoder framework that generates sharp and natural fused images. Furthermore, We pioneer the extension of image fusion techniques from static images to the video domain while preserving temporal consistency across frames, thus advancing the perceptual capabilities required for autonomous navigation. Extensive experiments on public datasets demonstrate that DAE-Fuse achieves state-of-the-art performance on multiple benchmarks, with superior generalizability to tasks like medical image fusion.
Generative models can create entirely new images, but they can also partially modify real images in ways that are undetectable to the human eye. In this paper, we address the challenge of automatically detecting such local manipulations. One of the most pressing problems in deepfake detection remains the ability of models to generalize to different classes of generators. In the case of fully manipulated images, representations extracted from large self-supervised models (such as CLIP) provide a promising direction towards more robust detectors. Here, we introduce DeCLIP, a first attempt to leverage such large pretrained features for detecting local manipulations. We show that, when combined with a reasonably large convolutional decoder, pretrained self-supervised representations are able to perform localization and improve generalization capabilities over existing methods. Unlike previous work, our approach is able to perform localization on the challenging case of latent diffusion models, where the entire image is affected by the fingerprint of the generator. Moreover, we observe that this type of data, which combines local semantic information with a global fingerprint, provides more stable generalization than other categories of generative methods.
Most of the prevalent approaches in speech prosody modeling rely on learning global style representations in a continuous latent space which encode and transfer the attributes of reference speech. However, recent work on neural codecs which are based on Residual Vector Quantization (RVQ) already shows great potential offering distinct advantages. We investigate the prosody modeling capabilities of the discrete space of such an RVQ-VAE model, modifying it to operate on the phoneme-level. We condition both the encoder and decoder of the model on linguistic representations and apply a global speaker embedding in order to factor out both phonetic and speaker information. We conduct an extensive set of investigations based on subjective experiments and objective measures to show that the phoneme-level discrete latent representations obtained this way achieves a high degree of disentanglement, capturing fine-grained prosodic information that is robust and transferable. The latent space turns out to have interpretable structure with its principal components corresponding to pitch and energy.
RT-DETR is the first real-time end-to-end transformer-based object detector. Its efficiency comes from the framework design and the Hungarian matching. However, compared to dense supervision detectors like the YOLO series, the Hungarian matching provides much sparser supervision, leading to insufficient model training and difficult to achieve optimal results. To address these issues, we proposed a hierarchical dense positive supervision method based on RT-DETR, named RT-DETRv3. Firstly, we introduce a CNN-based auxiliary branch that provides dense supervision that collaborates with the original decoder to enhance the encoder feature representation. Secondly, to address insufficient decoder training, we propose a novel learning strategy involving self-attention perturbation. This strategy diversifies label assignment for positive samples across multiple query groups, thereby enriching positive supervisions. Additionally, we introduce a shared-weight decoder branch for dense positive supervision to ensure more high-quality queries matching each ground truth. Notably, all aforementioned modules are training-only. We conduct extensive experiments to demonstrate the effectiveness of our approach on COCO val2017. RT-DETRv3 significantly outperforms existing real-time detectors, including the RT-DETR series and the YOLO series. For example, RT-DETRv3-R18 achieves 48.1% AP (+1.6%/+1.4%) compared to RT-DETR-R18/RT-DETRv2-R18, while maintaining the same latency. Furthermore, RT-DETRv3-R101 can attain an impressive 54.6% AP outperforming YOLOv10-X. The code will be released at https://github.com/clxia12/RT-DETRv3.
Vision Transformers (ViTs) have emerged as the backbone of many segmentation models, consistently achieving state-of-the-art (SOTA) performance. However, their success comes at a significant computational cost. Image token pruning is one of the most effective strategies to address this complexity. However, previous approaches fall short when applied to more complex task-oriented segmentation (TOS), where the class of each image patch is not predefined but dependent on the specific input task. This work introduces the Vision Language Guided Token Pruning (VLTP), a novel token pruning mechanism that can accelerate ViT-based segmentation models, particularly for TOS guided by multi-modal large language model (MLLM). We argue that ViT does not need to process every image token through all of its layers -- only the tokens related to reasoning tasks are necessary. We design a new pruning decoder to take both image tokens and vision-language guidance as input to predict the relevance of each image token to the task. Only image tokens with high relevance are passed to deeper layers of the ViT. Experiments show that the VLTP framework reduces the computational costs of ViT by approximately 25% without performance degradation and by around 40% with only a 1% performance drop. The code associated with this study can be found at this URL.
We introduce VistaFormer, a lightweight Transformer-based model architecture for the semantic segmentation of remote-sensing images. This model uses a multi-scale Transformer-based encoder with a lightweight decoder that aggregates global and local attention captured in the encoder blocks. VistaFormer uses position-free self-attention layers which simplifies the model architecture and removes the need to interpolate temporal and spatial codes, which can reduce model performance when training and testing image resolutions differ. We investigate simple techniques for filtering noisy input signals like clouds and demonstrate that improved model scalability can be achieved by substituting Multi-Head Self-Attention (MHSA) with Neighbourhood Attention (NA). Experiments on the PASTIS and MTLCC crop-type segmentation benchmarks show that VistaFormer achieves better performance than comparable models and requires only 8% of the floating point operations using MHSA and 11% using NA while also using fewer trainable parameters. VistaFormer with MHSA improves on state-of-the-art mIoU scores by 0.1% on the PASTIS benchmark and 3% on the MTLCC benchmark while VistaFormer with NA improves on the MTLCC benchmark by 3.7%.
In this paper, we introduce a novel general speech restoration model: the Dual-path Magnitude (DM) network, designed to address multiple distortions including noise, reverberation, and bandwidth degradation effectively. The DM network employs dual parallel magnitude decoders that share parameters: one uses a masking-based algorithm for distortion removal and the other employs a mapping-based approach for speech restoration. A novel aspect of the DM network is the integration of the magnitude spectrogram output from the masking decoder into the mapping decoder through a skip connection, enhancing the overall restoration capability. This integrated approach overcomes the inherent limitations observed in previous models, as detailed in a step-by-step analysis. The experimental results demonstrate that the DM network outperforms other baseline models in the comprehensive aspect of general speech restoration, achieving substantial restoration with fewer parameters.
Neural networks are typically black-boxes that remain opaque with regards to their decision mechanisms. Several works in the literature have proposed post-hoc explanation methods to alleviate this issue. This paper proposes LMAC-TD, a post-hoc explanation method that trains a decoder to produce explanations directly in the time domain. This methodology builds upon the foundation of L-MAC, Listenable Maps for Audio Classifiers, a method that produces faithful and listenable explanations. We incorporate SepFormer, a popular transformer-based time-domain source separation architecture. We show through a user study that LMAC-TD significantly improves the audio quality of the produced explanations while not sacrificing from faithfulness.
<div><p>Diffusion MRI is a powerful tool that serves as a bridge between brain microstructure and cognition. Recent advancements in cognitive neuroscience have highlighted the persistent challenge of understanding how individual differences in brain structure influence behavior, especially in healthy people. While traditional linear models like Canonical Correlation Analysis (CCA) and Partial Least Squares (PLS) have been fundamental in this analysis, they face limitations, particularly with high-dimensional data analysis outside the training sample. To address these issues, we introduce a novel approach using deep learninga multivariate autoencoder model-to explore the complex non-linear relationships between brain microstructure and cognitive functions. The model's architecture involves separate encoder modules for brain structure and cognitive data, with a shared decoder, facilitating the analysis of multivariate patterns across these domains. Both encoders were trained simultaneously, before the decoder, to ensure a good latent representation that captures the phenomenon. Using data from the Human Connectome Project, our study centres on the insula's role in cognitive processes. Through rigorous validation, including 5 sample analyses for out-of-sample analysis, our results demonstrate that the multivariate autoencoder model outperforms traditional methods in capturing and generalizing correlations between brain and behavior beyond the training sample. These findings underscore the potential of deep learning models to enhance our understanding of brain-behavior relationships in cognitive neuroscience, offering more accurate and comprehensive insights despite the complexities inherent in neuroimaging studies.</p></div>
Recent advances in deep learning-based joint source-channel coding (DJSCC) have shown promise for end-to-end semantic image transmission. However, most existing schemes primarily focus on optimizing pixel-wise metrics, which often fail to align with human perception, leading to lower perceptual quality. In this letter, we propose a novel generative DJSCC approach using conditional diffusion models to enhance the perceptual quality of transmitted images. Specifically, by utilizing entropy models, we effectively manage transmission bandwidth based on the estimated entropy of transmitted sym-bols. These symbols are then used at the receiver as conditional information to guide a conditional diffusion decoder in image reconstruction. Our model is built upon the emerging advanced mamba-like linear attention (MLLA) skeleton, which excels in image processing tasks while also offering fast inference speed. Besides, we introduce a multi-stage training strategy to ensure the stability and improve the overall performance of the model. Simulation results demonstrate that our proposed method significantly outperforms existing approaches in terms of perceptual quality.
Survival analysis using pathology images poses a considerable challenge, as it requires the localization of relevant information from the multitude of tiles within whole slide images (WSIs). Current methods typically resort to a two-stage approach, where a pre-trained network extracts features from tiles, which are then used by survival models. This process, however, does not optimize the survival models in an end-to-end manner, and the pre-extracted features may not be ideally suited for survival prediction. To address this limitation, we present a novel end-to-end Visual Prompt Tuning framework for survival analysis, named VPTSurv. VPTSurv refines feature embeddings through an efficient encoder-decoder framework. The encoder remains fixed while the framework introduces tunable visual prompts and adaptors, thus permitting end-to-end training specifically for survival prediction by optimizing only the lightweight adaptors and the decoder. Moreover, the versatile VPTSurv framework accommodates multi-source information as prompts, thereby enriching the survival model. VPTSurv achieves substantial increases of 8.7% and 12.5% in the C-index on two immunohistochemical pathology image datasets. These significant improvements highlight the transformative potential of the end-to-end VPT framework over traditional two-stage methods.
Large language models (LLMs) have become proficient at solving a wide variety of tasks, including those involving multi-modal inputs. In particular, instantiating an LLM (such as LLaMA) with a speech encoder and training it on paired data imparts speech recognition (ASR) abilities to the decoder-only model, hence called Speech-LLaMA. Nevertheless, due to the sequential nature of auto-regressive inference and the relatively large decoder, Speech-LLaMA models require relatively high inference time. In this work, we propose to speed up Speech-LLaMA inference by predicting multiple tokens in the same decoding step. We explore several model architectures that enable this, and investigate their performance using threshold-based and verification-based inference strategies. We also propose a prefix-based beam search decoding method that allows efficient minimum word error rate (MWER) training for such models. We evaluate our models on a variety of public benchmarks, where they reduce the number of decoder calls by ~3.2x while maintaining or improving WER performance.
In this paper, we introduce SSR-Speech, a neural codec autoregressive model designed for stable, safe, and robust zero-shot textbased speech editing and text-to-speech synthesis. SSR-Speech is built on a Transformer decoder and incorporates classifier-free guidance to enhance the stability of the generation process. A watermark Encodec is proposed to embed frame-level watermarks into the edited regions of the speech so that which parts were edited can be detected. In addition, the waveform reconstruction leverages the original unedited speech segments, providing superior recovery compared to the Encodec model. Our approach achieves state-of-the-art performance in the RealEdit speech editing task and the LibriTTS text-to-speech task, surpassing previous methods. Furthermore, SSR-Speech excels in multi-span speech editing and also demonstrates remarkable robustness to background sounds. The source code and demos are released.
Continuous speech can be converted into a discrete sequence by deriving discrete units from the hidden features of self-supervised learned (SSL) speech models. Although SSL models are becoming larger and trained on more data, they are often sensitive to real-life distortions like additive noise or reverberation, which translates to a shift in discrete units. We propose a parameter-efficient approach to generate noise-robust discrete units from pre-trained SSL models by training a small encoder-decoder model, with or without adapters, to simultaneously denoise and discretise the hidden features of the SSL model. The model learns to generate a clean discrete sequence for a noisy utterance, conditioned on the SSL features. The proposed denoiser outperforms several pre-training methods on the tasks of noisy discretisation and noisy speech recognition, and can be finetuned to the target environment with a few recordings of unlabeled target data.
To overcome the burden on the memory size and bandwidth due to ever-increasing size of large language models (LLMs), aggressive weight quantization has been recently studied, while lacking research on quantizing activations. In this paper, we present a hardware-software co-design method that results in an energy-efficient LLM accelerator, named OPAL, for generation tasks. First of all, a novel activation quantization method that leverages the microscaling data format while preserving several outliers per sub-tensor block (e.g., four out of 128 elements) is proposed. Second, on top of preserving outliers, mixed precision is utilized that sets 5-bit for inputs to sensitive layers in the decoder block of an LLM, while keeping inputs to less sensitive layers to 3-bit. Finally, we present the OPAL hardware architecture that consists of FP units for handling outliers and vectorized INT multipliers for dominant non-outlier related operations. In addition, OPAL uses log2-based approximation on softmax operations that only requires shift and subtraction to maximize power efficiency. As a result, we are able to improve the energy efficiency by 1.6~2.2x, and reduce the area by 2.4~3.1x with negligible accuracy loss, i.e., <1 perplexity increase.
Both manual (relating to the use of hands) and non-manual markers (NMM), such as facial expressions or mouthing cues, are important for providing the complete meaning of phrases in American Sign Language (ASL). Efforts have been made in advancing sign language to spoken/written language understanding, but most of these have primarily focused on manual features only. In this work, using advanced neural machine translation methods, we examine and report on the extent to which facial expressions contribute to understanding sign language phrases. We present a sign language translation architecture consisting of two-stream encoders, with one encoder handling the face and the other handling the upper body (with hands). We propose a new parallel cross-attention decoding mechanism that is useful for quantifying the influence of each input modality on the output. The two streams from the encoder are directed simultaneously to different attention stacks in the decoder. Examining the properties of the parallel cross-attention weights allows us to analyze the importance of facial markers compared to body and hand features during a translating task.
Depth completion aims to predict dense depth maps with sparse depth measurements from a depth sensor. Currently, Convolutional Neural Network (CNN) based models are the most popular methods applied to depth completion tasks. However, despite the excellent high-end performance, they suffer from a limited representation area. To overcome the drawbacks of CNNs, a more effective and powerful method has been presented: the Transformer, which is an adaptive self-attention setting sequence-to-sequence model. While the standard Transformer quadratically increases the computational cost from the key-query dot-product of input resolution which improperly employs depth completion tasks. In this work, we propose a different window-based Transformer architecture for depth completion tasks named Sparse-to-Dense Transformer (SDformer). The network consists of an input module for the depth map and RGB image features extraction and concatenation, a U-shaped encoder-decoder Transformer for extracting deep features, and a refinement module. Specifically, we first concatenate the depth map features with the RGB image features through the input model. Then, instead of calculating self-attention with the whole feature maps, we apply different window sizes to extract the long-range depth dependencies. Finally, we refine the predicted features from the input module and the U-shaped encoder-decoder Transformer module to get the enriching depth features and employ a convolution layer to obtain the dense depth map. In practice, the SDformer obtains state-of-the-art results against the CNN-based depth completion models with lower computing loads and parameters on the NYU Depth V2 and KITTI DC datasets.