DT-JRD: Deep Transformer based Just Recognizable Difference Prediction Model for Video Coding for Machines

Just Recognizable Difference (JRD) represents the minimum visual difference that is detectable by machine vision, which can be exploited to promote machine vision oriented visual signal processing. In this paper, we propose a Deep Transformer based JRD (DT-JRD) prediction model for Video Coding for Machines (VCM), where the accurately predicted JRD can be used reduce the coding bit rate while maintaining the accuracy of machine tasks. Firstly, we model the JRD prediction as a multi-class classification and propose a DT-JRD prediction model that integrates an improved embedding, a content and distortion feature extraction, a multi-class classification and a novel learning strategy. Secondly, inspired by the perception property that machine vision exhibits a similar response to distortions near JRD, we propose an asymptotic JRD loss by using Gaussian Distribution-based Soft Labels (GDSL), which significantly extends the number of training labels and relaxes classification boundaries. Finally, we propose a DT-JRD based VCM to reduce the coding bits while maintaining the accuracy of object detection. Extensive experimental results demonstrate that the mean absolute error of the predicted JRD by the DT-JRD is 5.574, outperforming the state-of-the-art JRD prediction model by 13.1%. Coding experiments shows that comparing with the VVC, the DT-JRD based VCM achieves an average of 29.58% bit rate reduction while maintaining the object detection accuracy.

Joint multi-dimensional dynamic attention and transformer for general image restoration

Outdoor images often suffer from severe degradation due to rain, haze, and noise, impairing image quality and challenging high-level tasks. Current image restoration methods struggle to handle complex degradation while maintaining efficiency. This paper introduces a novel image restoration architecture that combines multi-dimensional dynamic attention and self-attention within a U-Net framework. To leverage the global modeling capabilities of transformers and the local modeling capabilities of convolutions, we integrate sole CNNs in the encoder-decoder and sole transformers in the latent layer. Additionally, we design convolutional kernels with selected multi-dimensional dynamic attention to capture diverse degraded inputs efficiently. A transformer block with transposed self-attention further enhances global feature extraction while maintaining efficiency. Extensive experiments demonstrate that our method achieves a better balance between performance and computational complexity across five image restoration tasks: deraining, deblurring, denoising, dehazing, and enhancement, as well as superior performance for high-level vision tasks. The source code will be available at https://github.com/House-yuyu/MDDA-former.

LFIC-DRASC: Deep Light Field Image Compression Using Disentangled Representation and Asymmetrical Strip Convolution

Light-Field (LF) image is emerging 4D data of light rays that is capable of realistically presenting spatial and angular information of 3D scene. However, the large data volume of LF images becomes the most challenging issue in real-time processing, transmission, and storage. In this paper, we propose an end-to-end deep LF Image Compression method Using Disentangled Representation and Asymmetrical Strip Convolution (LFIC-DRASC) to improve coding efficiency. Firstly, we formulate the LF image compression problem as learning a disentangled LF representation network and an image encoding-decoding network. Secondly, we propose two novel feature extractors that leverage the structural prior of LF data by integrating features across different dimensions. Meanwhile, disentangled LF representation network is proposed to enhance the LF feature disentangling and decoupling. Thirdly, we propose the LFIC-DRASC for LF image compression, where two Asymmetrical Strip Convolution (ASC) operators, i.e. horizontal and vertical, are proposed to capture long-range correlation in LF feature space. These two ASC operators can be combined with the square convolution to further decouple LF features, which enhances the model ability in representing intricate spatial relationships. Experimental results demonstrate that the proposed LFIC-DRASC achieves an average of 20.5\% bit rate reductions comparing with the state-of-the-art methods.

PC-SRIF: Preconditioned Cholesky-based Square Root Information Filter for Vision-aided Inertial Navigation

In this paper, we introduce a novel estimator for vision-aided inertial navigation systems (VINS), the Preconditioned Cholesky-based Square Root Information Filter (PC-SRIF). When solving linear systems, employing Cholesky decomposition offers superior efficiency but can compromise numerical stability. Due to this, existing VINS utilizing (Square Root) Information Filters often opt for QR decomposition on platforms where single precision is preferred, avoiding the numerical challenges associated with Cholesky decomposition. While these issues are often attributed to the ill-conditioned information matrix in VINS, our analysis reveals that this is not an inherent property of VINS but rather a consequence of specific parameterizations. We identify several factors that contribute to an ill-conditioned information matrix and propose a preconditioning technique to mitigate these conditioning issues. Building on this analysis, we present PC-SRIF, which exhibits remarkable stability in performing Cholesky decomposition in single precision when solving linear systems in VINS. Consequently, PC-SRIF achieves superior theoretical efficiency compared to alternative estimators. To validate the efficiency advantages and numerical stability of PC-SRIF based VINS, we have conducted well controlled experiments, which provide empirical evidence in support of our theoretical findings. Remarkably, in our VINS implementation, PC-SRIF's runtime is 41% faster than QR-based SRIF.

Global-Local Progressive Integration Network for Blind Image Quality Assessment

Vision transformers (ViTs) excel in computer vision for modeling long-term dependencies, yet face two key challenges for image quality assessment (IQA): discarding fine details during patch embedding, and requiring extensive training data due to lack of inductive biases. In this study, we propose a Global-Local progressive INTegration network for IQA, called GlintIQA, to address these issues through three key components: 1) Hybrid feature extraction combines ViT-based global feature extractor (VGFE) and convolutional neural networks (CNNs)-based local feature extractor (CLFE) to capture global coarse-grained features and local fine-grained features, respectively. The incorporation of CNNs mitigates the patch-level information loss and inductive bias constraints inherent to ViT architectures. 2) Progressive feature integration leverages diverse kernel sizes in embedding to spatially align coarse- and fine-grained features, and progressively aggregate these features by interactively stacking channel-wise attention and spatial enhancement modules to build effective quality-aware representations. 3) Content similarity-based labeling approach is proposed that automatically assigns quality labels to images with diverse content based on subjective quality scores. This addresses the scarcity of labeled training data in synthetic datasets and bolsters model generalization. The experimental results demonstrate the efficacy of our approach, yielding 5.04% average SROCC gains on cross-authentic dataset evaluations. Moreover, our model and its counterpart pre-trained on the proposed dataset respectively exhibited 5.40% and 13.23% improvements on across-synthetic datasets evaluation. The codes and proposed dataset will be released at https://github.com/XiaoqiWang/GlintIQA.

Multi-Granularity Semantic Revision for Large Language Model Distillation

Knowledge distillation plays a key role in compressing the Large Language Models (LLMs), which boosts a small-size student model under large teacher models' guidance. However, existing LLM distillation methods overly rely on student-generated outputs, which may introduce generation errors and misguide the distillation process. Moreover, the distillation loss functions introduced in previous art struggle to align the most informative part due to the complex distribution of LLMs' outputs. To address these problems, we propose a multi-granularity semantic revision method for LLM distillation. At the sequence level, we propose a sequence correction and re-generation (SCRG) strategy. SCRG first calculates the semantic cognitive difference between the teacher and student to detect the error token, then corrects it with the teacher-generated one, and re-generates the sequence to reduce generation errors and enhance generation diversity. At the token level, we design a distribution adaptive clipping Kullback-Leibler (DAC-KL) loss as the distillation objective function. DAC-KL loss exploits a learnable sub-network to adaptively extract semantically dense areas from the teacher's output, avoiding the interference of redundant information in the distillation process. Finally, at the span level, we leverage the span priors of a sequence to compute the probability correlations within spans, and constrain the teacher and student's probability correlations to be consistent, further enhancing the transfer of semantic information. Extensive experiments across different model families with parameters ranging from 0.1B to 13B demonstrate the superiority of our method compared to existing methods.

The Ninth NTIRE 2024 Efficient Super-Resolution Challenge Report

This paper provides a comprehensive review of the NTIRE 2024 challenge, focusing on efficient single-image super-resolution (ESR) solutions and their outcomes. The task of this challenge is to super-resolve an input image with a magnification factor of x4 based on pairs of low and corresponding high-resolution images. The primary objective is to develop networks that optimize various aspects such as runtime, parameters, and FLOPs, while still maintaining a peak signal-to-noise ratio (PSNR) of approximately 26.90 dB on the DIV2K_LSDIR_valid dataset and 26.99 dB on the DIV2K_LSDIR_test dataset. In addition, this challenge has 4 tracks including the main track (overall performance), sub-track 1 (runtime), sub-track 2 (FLOPs), and sub-track 3 (parameters). In the main track, all three metrics (ie runtime, FLOPs, and parameter count) were considered. The ranking of the main track is calculated based on a weighted sum-up of the scores of all other sub-tracks. In sub-track 1, the practical runtime performance of the submissions was evaluated, and the corresponding score was used to determine the ranking. In sub-track 2, the number of FLOPs was considered. The score calculated based on the corresponding FLOPs was used to determine the ranking. In sub-track 3, the number of parameters was considered. The score calculated based on the corresponding parameters was used to determine the ranking. RLFN is set as the baseline for efficiency measurement. The challenge had 262 registered participants, and 34 teams made valid submissions. They gauge the state-of-the-art in efficient single-image super-resolution. To facilitate the reproducibility of the challenge and enable other researchers to build upon these findings, the code and the pre-trained model of validated solutions are made publicly available at https://github.com/Amazingren/NTIRE2024_ESR/.

Knowledge Distillation with Multi-granularity Mixture of Priors for Image Super-Resolution

Knowledge distillation (KD) is a promising yet challenging model compression technique that transfers rich learning representations from a well-performing but cumbersome teacher model to a compact student model. Previous methods for image super-resolution (SR) mostly compare the feature maps directly or after standardizing the dimensions with basic algebraic operations (e.g. average, dot-product). However, the intrinsic semantic differences among feature maps are overlooked, which are caused by the disparate expressive capacity between the networks. This work presents MiPKD, a multi-granularity mixture of prior KD framework, to facilitate efficient SR model through the feature mixture in a unified latent space and stochastic network block mixture. Extensive experiments demonstrate the effectiveness of the proposed MiPKD method.

Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement

Adversarial training (AT) is currently one of the most effective ways to obtain the robustness of deep neural networks against adversarial attacks. However, most AT methods suffer from robust overfitting, i.e., a significant generalization gap in adversarial robustness between the training and testing curves. In this paper, we first identify a connection between robust overfitting and the excessive memorization of noisy labels in AT from a view of gradient norm. As such label noise is mainly caused by a distribution mismatch and improper label assignments, we are motivated to propose a label refinement approach for AT. Specifically, our Self-Guided Label Refinement first self-refines a more accurate and informative label distribution from over-confident hard labels, and then it calibrates the training by dynamically incorporating knowledge from self-distilled models into the current model and thus requiring no external teachers. Empirical results demonstrate that our method can simultaneously boost the standard accuracy and robust performance across multiple benchmark datasets, attack types, and architectures. In addition, we also provide a set of analyses from the perspectives of information theory to dive into our method and suggest the importance of soft labels for robust generalization.

Eliminating Warping Shakes for Unsupervised Online Video Stitching

In this paper, we retarget video stitching to an emerging issue, named warping shake, when extending image stitching to video stitching. It unveils the temporal instability of warped content in non-overlapping regions, despite image stitching having endeavored to preserve the natural structures. Therefore, in most cases, even if the input videos to be stitched are stable, the stitched video will inevitably cause undesired warping shakes and affect the visual experience. To eliminate the shakes, we propose StabStitch to simultaneously realize video stitching and video stabilization in a unified unsupervised learning framework. Starting from the camera paths in video stabilization, we first derive the expression of stitching trajectories in video stitching by elaborately integrating spatial and temporal warps. Then a warp smoothing model is presented to optimize them with a comprehensive consideration regarding content alignment, trajectory smoothness, spatial consistency, and online collaboration. To establish an evaluation benchmark and train the learning framework, we build a video stitching dataset with a rich diversity in camera motions and scenes. Compared with existing stitching solutions, StabStitch exhibits significant superiority in scene robustness and inference speed in addition to stitching and stabilization performance, contributing to a robust and real-time online video stitching system. The code and dataset will be available at https://github.com/nie-lang/StabStitch.