While replacing Gaussian decoders with a conditional diffusion model enhances the perceptual quality of reconstructions in neural image compression, their lack of inductive bias for image data restricts their ability to achieve state-of-the-art perceptual levels. To address this limitation, we adopt a non-isotropic diffusion model at the decoder side. This model imposes an inductive bias aimed at distinguishing between frequency contents, thereby facilitating the generation of high-quality images. Moreover, our framework is equipped with a novel entropy model that accurately models the probability distribution of latent representation by exploiting spatio-channel correlations in latent space, while accelerating the entropy decoding step. This channel-wise entropy model leverages both local and global spatial contexts within each channel chunk. The global spatial context is built upon the Transformer, which is specifically designed for image compression tasks. The designed Transformer employs a Laplacian-shaped positional encoding, the learnable parameters of which are adaptively adjusted for each channel cluster. Our experiments demonstrate that our proposed framework yields better perceptual quality compared to cutting-edge generative-based codecs, and the proposed entropy model contributes to notable bitrate savings.
The one-shot talking-head generation learns to synthesize a talking-head video with one source portrait image under the driving of same or different identity video. Usually these methods require plane-based pixel transformations via Jacobin matrices or facial image warps for novel poses generation. The constraints of using a single image source and pixel displacements often compromise the clarity of the synthesized images. Some methods try to improve the quality of synthesized videos by introducing additional super-resolution modules, but this will undoubtedly increase computational consumption and destroy the original data distribution. In this work, we propose an adaptive high-quality talking-head video generation method, which synthesizes high-resolution video without additional pre-trained modules. Specifically, inspired by existing super-resolution methods, we down-sample the one-shot source image, and then adaptively reconstruct high-frequency details via an encoder-decoder module, resulting in enhanced video clarity. Our method consistently improves the quality of generated videos through a straightforward yet effective strategy, substantiated by quantitative and qualitative evaluations. The code and demo video are available on: \url{https://github.com/Songluchuan/AdaSR-TalkingHead/}.
Auditory attention decoding (AAD) aims to extract from brain activity the attended speaker amidst candidate speakers, offering promising applications for neuro-steered hearing devices and brain-computer interfacing. This pilot study makes a first step towards AAD using the noise-tagging stimulus protocol, which evokes reliable code-modulated evoked potentials, but is minimally explored in the auditory modality. Participants were sequentially presented with two Dutch speech stimuli that were amplitude-modulated with a unique binary pseudo-random noise-code, effectively tagging these with additional decodable information. We compared the decoding of unmodulated audio against audio modulated with various modulation depths, and a conventional AAD method against a standard method to decode noise-codes. Our pilot study revealed higher performances for the conventional method with 70 to 100 percent modulation depths compared to unmodulated audio. The noise-code decoder did not further improve these results. These fundamental insights highlight the potential of integrating noise-codes in speech to enhance auditory speaker detection when multiple speakers are presented simultaneously.
Empathetic response generation endeavors to empower dialogue systems to perceive speakers' emotions and generate empathetic responses accordingly. Psychological research demonstrates that emotion, as an essential factor in empathy, encompasses trait emotions, which are static and context-independent, and state emotions, which are dynamic and context-dependent. However, previous studies treat them in isolation, leading to insufficient emotional perception of the context, and subsequently, less effective empathetic expression. To address this problem, we propose Combining Trait and State emotions for Empathetic Response Model (CTSM). Specifically, to sufficiently perceive emotions in dialogue, we first construct and encode trait and state emotion embeddings, and then we further enhance emotional perception capability through an emotion guidance module that guides emotion representation. In addition, we propose a cross-contrastive learning decoder to enhance the model's empathetic expression capability by aligning trait and state emotions between generated responses and contexts. Both automatic and manual evaluation results demonstrate that CTSM outperforms state-of-the-art baselines and can generate more empathetic responses. Our code is available at https://github.com/wangyufeng-empty/CTSM
Representation learning (RL) methods for cyberattack detection face the diversity and sophistication of attack data, leading to the issue of mixed representations of different classes, particularly as the number of classes increases. To address this, the paper proposes a novel deep learning architecture/model called the Twin Auto-Encoder (TAE). TAE first maps the input data into latent space and then deterministically shifts data samples of different classes further apart to create separable data representations, referred to as representation targets. TAE's decoder then projects the input data into these representation targets. After training, TAE's decoder extracts data representations. TAE's representation target serves as a novel dynamic codeword, which refers to the vector that represents a specific class. This vector is updated after each training epoch for every data sample, in contrast to the conventional fixed codeword that does not incorporate information from the input data. We conduct extensive experiments on diverse cybersecurity datasets, including seven IoT botnet datasets, two network IDS datasets, three malware datasets, one cloud DDoS dataset, and ten artificial datasets as the number of classes increases. TAE boosts accuracy and F-score in attack detection by around 2% compared to state-of-the-art models, achieving up to 96.1% average accuracy in IoT attack detection. Additionally, TAE is well-suited for cybersecurity applications and potentially for IoT systems, with a model size of approximately 1 MB and an average running time of around 2.6E-07 seconds for extracting a data sample.
Point scene understanding is a challenging task to process real-world scene point cloud, which aims at segmenting each object, estimating its pose, and reconstructing its mesh simultaneously. Recent state-of-the-art method first segments each object and then processes them independently with multiple stages for the different sub-tasks. This leads to a complex pipeline to optimize and makes it hard to leverage the relationship constraints between multiple objects. In this work, we propose a novel Disentangled Object-Centric TRansformer (DOCTR) that explores object-centric representation to facilitate learning with multiple objects for the multiple sub-tasks in a unified manner. Each object is represented as a query, and a Transformer decoder is adapted to iteratively optimize all the queries involving their relationship. In particular, we introduce a semantic-geometry disentangled query (SGDQ) design that enables the query features to attend separately to semantic information and geometric information relevant to the corresponding sub-tasks. A hybrid bipartite matching module is employed to well use the supervisions from all the sub-tasks during training. Qualitative and quantitative experimental results demonstrate that our method achieves state-of-the-art performance on the challenging ScanNet dataset. Code is available at https://github.com/SAITPublic/DOCTR.
Ultrasound imaging is crucial for evaluating organ morphology and function, yet depth adjustment can degrade image quality and field-of-view, presenting a depth-dependent dilemma. Traditional interpolation-based zoom-in techniques often sacrifice detail and introduce artifacts. Motivated by the potential of arbitrary-scale super-resolution to naturally address these inherent challenges, we present the Residual Dense Swin Transformer Network (RDSTN), designed to capture the non-local characteristics and long-range dependencies intrinsic to ultrasound images. It comprises a linear embedding module for feature enhancement, an encoder with shifted-window attention for modeling non-locality, and an MLP decoder for continuous detail reconstruction. This strategy streamlines balancing image quality and field-of-view, which offers superior textures over traditional methods. Experimentally, RDSTN outperforms existing approaches while requiring fewer parameters. In conclusion, RDSTN shows promising potential for ultrasound image enhancement by overcoming the limitations of conventional interpolation-based methods and achieving depth-independent imaging.
Image denoising is a fundamental task in computer vision. While prevailing deep learning-based supervised and self-supervised methods have excelled in eliminating in-distribution noise, their susceptibility to out-of-distribution (OOD) noise remains a significant challenge. The recent emergence of contrastive language-image pre-training (CLIP) model has showcased exceptional capabilities in open-world image recognition and segmentation. Yet, the potential for leveraging CLIP to enhance the robustness of low-level tasks remains largely unexplored. This paper uncovers that certain dense features extracted from the frozen ResNet image encoder of CLIP exhibit distortion-invariant and content-related properties, which are highly desirable for generalizable denoising. Leveraging these properties, we devise an asymmetrical encoder-decoder denoising network, which incorporates dense features including the noisy image and its multi-scale features from the frozen ResNet encoder of CLIP into a learnable image decoder to achieve generalizable denoising. The progressive feature augmentation strategy is further proposed to mitigate feature overfitting and improve the robustness of the learnable decoder. Extensive experiments and comparisons conducted across diverse OOD noises, including synthetic noise, real-world sRGB noise, and low-dose CT image noise, demonstrate the superior generalization ability of our method.
3D instance segmentation (3DIS) is a crucial task, but point-level annotations are tedious in fully supervised settings. Thus, using bounding boxes (bboxes) as annotations has shown great potential. The current mainstream approach is a two-step process, involving the generation of pseudo-labels from box annotations and the training of a 3DIS network with the pseudo-labels. However, due to the presence of intersections among bboxes, not every point has a determined instance label, especially in overlapping areas. To generate higher quality pseudo-labels and achieve more precise weakly supervised 3DIS results, we propose the Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation (BSNet), which devises a novel pseudo-labeler called Simulation-assisted Transformer. The labeler consists of two main components. The first is Simulation-assisted Mean Teacher, which introduces Mean Teacher for the first time in this task and constructs simulated samples to assist the labeler in acquiring prior knowledge about overlapping areas. To better model local-global structure, we also propose Local-Global Aware Attention as the decoder for teacher and student labelers. Extensive experiments conducted on the ScanNetV2 and S3DIS datasets verify the superiority of our designs. Code is available at \href{https://github.com/peoplelu/BSNet}{https://github.com/peoplelu/BSNet}.
End-to-End Neural Diarization with Encoder-Decoder based Attractor (EEND-EDA) is an end-to-end neural model for automatic speaker segmentation and labeling. It achieves the capability to handle flexible number of speakers by estimating the number of attractors. EEND-EDA, however, struggles to accurately capture local speaker dynamics. This work proposes an auxiliary loss that aims to guide the Transformer encoders at the lower layer of EEND-EDA model to enhance the effect of self-attention modules using speaker activity information. The results evaluated on public dataset Mini LibriSpeech, demonstrates the effectiveness of the work, reducing Diarization Error Rate from 30.95% to 28.17%. We will release the source code on GitHub to allow further research and reproducibility.
Interactions with virtual assistants typically start with a predefined trigger phrase followed by the user command. To make interactions with the assistant more intuitive, we explore whether it is feasible to drop the requirement that users must begin each command with a trigger phrase. We explore this task in three ways: First, we train classifiers using only acoustic information obtained from the audio waveform. Second, we take the decoder outputs of an automatic speech recognition (ASR) system, such as 1-best hypotheses, as input features to a large language model (LLM). Finally, we explore a multimodal system that combines acoustic and lexical features, as well as ASR decoder signals in an LLM. Using multimodal information yields relative equal-error-rate improvements over text-only and audio-only models of up to 39% and 61%. Increasing the size of the LLM and training with low-rank adaption leads to further relative EER reductions of up to 18% on our dataset.
Retrieval-augmented generation models augment knowledge encoded in a language model by providing additional relevant external knowledge (context) during generation. Although it has been shown that the quantity and quality of context impact the performance of retrieval-augmented generation models during inference, limited research explores how these characteristics affect model training. This paper explores how context quantity and quality during model training affect the performance of Fusion-in-Decoder (FiD), the state-of-the-art retrieval-augmented generation model, in extractive open-domain question answering tasks. Experimental results suggest that FiD models overfit to context quality during training and show suboptimal performance when evaluated on different context quality. Through the experimental results, we also reveal FiD models trained with different context quality have different cross-attention distribution patterns. Specifically, as context quality during training increases, FiD models tend to attend more uniformly to each passage in context. Finally, based on these observations, we propose a method to mitigate overfitting to specific context quality by introducing bias to the cross-attention distribution, which we demonstrate to be effective in improving the performance of FiD models on different context quality.
The availability of unprecedented unsupervised training data, along with neural scaling laws, has resulted in an unprecedented surge in model size and compute requirements for serving/training LLMs. However, the main performance bottleneck is increasingly shifting to memory bandwidth. Over the past 20 years, peak server hardware FLOPS has been scaling at 3.0x/2yrs, outpacing the growth of DRAM and interconnect bandwidth, which have only scaled at 1.6 and 1.4 times every 2 years, respectively. This disparity has made memory, rather than compute, the primary bottleneck in AI applications, particularly in serving. Here, we analyze encoder and decoder Transformer models and show how memory bandwidth can become the dominant bottleneck for decoder models. We argue for a redesign in model architecture, training, and deployment strategies to overcome this memory limitation.
Resource allocation in tactical ad-hoc networks presents unique challenges due to their dynamic and multi-hop nature. Accurate prediction of future network connectivity is essential for effective resource allocation in such environments. In this paper, we introduce the Spatial-Temporal Graph Encoder-Decoder (STGED) framework for Tactical Communication Networks that leverages both spatial and temporal features of network states to learn latent tactical behaviors effectively. STGED hierarchically utilizes graph-based attention mechanism to spatially encode a series of communication network states, leverages a recurrent neural network to temporally encode the evolution of states, and a fully-connected feed-forward network to decode the connectivity in the future state. Through extensive experiments, we demonstrate that STGED consistently outperforms baseline models by large margins across different time-steps input, achieving an accuracy of up to 99.2\% for the future state prediction task of tactical communication networks.
PSALM is a powerful extension of the Large Multi-modal Model (LMM) to address the segmentation task challenges. To overcome the limitation of the LMM being limited to textual output, PSALM incorporates a mask decoder and a well-designed input schema to handle a variety of segmentation tasks. This schema includes images, task instructions, conditional prompts, and mask tokens, which enable the model to generate and classify segmentation masks effectively. The flexible design of PSALM supports joint training across multiple datasets and tasks, leading to improved performance and task generalization. PSALM achieves superior results on several benchmarks, such as RefCOCO/RefCOCO+/RefCOCOg, COCO Panoptic Segmentation, and COCO-Interactive, and further exhibits zero-shot capabilities on unseen tasks, such as open-vocabulary segmentation, generalized referring expression segmentation and video object segmentation, making a significant step towards a GPT moment in computer vision. Through extensive experiments, PSALM demonstrates its potential to transform the domain of image segmentation, leveraging the robust visual understanding capabilities of LMMs as seen in natural language processing. Code and models are available at https://github.com/zamling/PSALM.
Visual relationship detection aims to identify objects and their relationships in images. Prior methods approach this task by adding separate relationship modules or decoders to existing object detection architectures. This separation increases complexity and hinders end-to-end training, which limits performance. We propose a simple and highly efficient decoder-free architecture for open-vocabulary visual relationship detection. Our model consists of a Transformer-based image encoder that represents objects as tokens and models their relationships implicitly. To extract relationship information, we introduce an attention mechanism that selects object pairs likely to form a relationship. We provide a single-stage recipe to train this model on a mixture of object and relationship detection data. Our approach achieves state-of-the-art relationship detection performance on Visual Genome and on the large-vocabulary GQA benchmark at real-time inference speeds. We provide ablations, real-world qualitative examples, and analyses of zero-shot performance.
In the context of high usability in single-class anomaly detection models, recent academic research has become concerned about the more complex multi-class anomaly detection. Although several papers have designed unified models for this task, they often overlook the utility of class labels, a potent tool for mitigating inter-class interference. To address this issue, we introduce a Multi-class Implicit Neural representation Transformer for unified Anomaly Detection (MINT-AD), which leverages the fine-grained category information in the training stage. By learning the multi-class distributions, the model generates class-aware query embeddings for the transformer decoder, mitigating inter-class interference within the reconstruction model. Utilizing such an implicit neural representation network, MINT-AD can project category and position information into a feature embedding space, further supervised by classification and prior probability loss functions. Experimental results on multiple datasets demonstrate that MINT-AD outperforms existing unified training models.
Human motion prediction is consisting in forecasting future body poses from historically observed sequences. It is a longstanding challenge due to motion's complex dynamics and uncertainty. Existing methods focus on building up complicated neural networks to model the motion dynamics. The predicted results are required to be strictly similar to the training samples with L2 loss in current training pipeline. However, little attention has been paid to the uncertainty property which is crucial to the prediction task. We argue that the recorded motion in training data could be an observation of possible future, rather than a predetermined result. In addition, existing works calculate the predicted error on each future frame equally during training, while recent work indicated that different frames could play different roles. In this work, a novel computationally efficient encoder-decoder model with uncertainty consideration is proposed, which could learn proper characteristics for future frames by a dynamic function. Experimental results on benchmark datasets demonstrate that our uncertainty consideration approach has obvious advantages both in quantity and quality. Moreover, the proposed method could produce motion sequences with much better quality that avoids the intractable shaking artefacts. We believe our work could provide a novel perspective to consider the uncertainty quality for the general motion prediction task and encourage the studies in this field. The code will be available in https://github.com/Motionpre/Adaptive-Salient-Loss-SAGGB.
Segment Anything Model~(SAM), a prompt-driven foundation model for natural image segmentation, has demonstrated impressive zero-shot performance. However, SAM does not work when directly applied to medical image segmentation tasks, since SAM lacks the functionality to predict semantic labels for predicted masks and needs to provide extra prompts, such as points or boxes, to segment target regions. Meanwhile, there is a huge gap between 2D natural images and 3D medical images, so the performance of SAM is imperfect for medical image segmentation tasks. Following the above issues, we propose MaskSAM, a novel mask classification prompt-free SAM adaptation framework for medical image segmentation. We design a prompt generator combined with the image encoder in SAM to generate a set of auxiliary classifier tokens, auxiliary binary masks, and auxiliary bounding boxes. Each pair of auxiliary mask and box prompts, which can solve the requirements of extra prompts, is associated with class label predictions by the sum of the auxiliary classifier token and the learnable global classifier tokens in the mask decoder of SAM to solve the predictions of semantic labels. Meanwhile, we design a 3D depth-convolution adapter for image embeddings and a 3D depth-MLP adapter for prompt embeddings. We inject one of them into each transformer block in the image encoder and mask decoder to enable pre-trained 2D SAM models to extract 3D information and adapt to 3D medical images. Our method achieves state-of-the-art performance on AMOS2022, 90.52% Dice, which improved by 2.7% compared to nnUNet. Our method surpasses nnUNet by 1.7% on ACDC and 1.0% on Synapse datasets.
Recent deblurring networks have effectively restored clear images from the blurred ones. However, they often struggle with generalization to unknown domains. Moreover, these models typically focus on distortion metrics such as PSNR and SSIM, neglecting the critical aspect of metrics aligned with human perception. To address these limitations, we propose DeblurDiNAT, a deblurring Transformer based on Dilated Neighborhood Attention. First, DeblurDiNAT employs an alternating dilation factor paradigm to capture both local and global blurred patterns, enhancing generalization and perceptual clarity. Second, a local cross-channel learner aids the Transformer block to understand the short-range relationships between adjacent channels. Additionally, we present a linear feed-forward network with a simple while effective design. Finally, a dual-stage feature fusion module is introduced as an alternative to the existing approach, which efficiently process multi-scale visual information across network levels. Compared to state-of-the-art models, our compact DeblurDiNAT demonstrates superior generalization capabilities and achieves remarkable performance in perceptual metrics, while maintaining a favorable model size.