Zhiying Jiang

Raphael Tang, Linqing Liu, Akshat Pandey et al.

Large-scale diffusion neural networks represent a substantial milestone in text-to-image generation, but they remain poorly understood, lacking interpretability analyses. In this paper, we perform a text-image attribution analysis on Stable Diffusion, a recently open-sourced model. To produce pixel-level attribution maps, we upscale and aggregate cross-attention word-pixel scores in the denoising subnetwork, naming our method DAAM. We evaluate its correctness by testing its semantic segmentation ability on nouns, as well as its generalized attribution quality on all parts of speech, rated by humans. We then apply DAAM to study the role of syntax in the pixel space, characterizing head--dependent heat map interaction patterns for ten common dependency relations. Finally, we study several semantic phenomena using DAAM, with a focus on feature entanglement, where we find that cohyponyms worsen generation quality and descriptive adjectives attend too broadly. To our knowledge, we are the first to interpret large diffusion models from a visuolinguistic perspective, which enables future lines of research. Our code is at https://github.com/castorini/daam.

Zhiying Jiang, Raihan Seraj, Marcos Villagra et al.

Training frontier-scale diffusion models often requires substantial computational resources concentrated in tightly-coupled clusters, limiting participation to well-resourced institutions. While Decentralized Diffusion Models (DDM) enable training multiple experts in isolation, existing approaches require 1176 GPU-days and homogeneous training objectives across all experts. We present an efficient framework that dramatically reduces resource requirements while supporting heterogeneous training objectives. Our approach combines three key contributions: (1) a heterogeneous decentralized training paradigm that allows experts to use different objectives (DDPM and Flow Matching), unified at inference time without any retraining; (2) pretrained checkpoint conversion from ImageNet-DDPM to Flow Matching objectives, accelerating convergence and enabling initialization without objective-specific pretraining; and (3) PixArt-$α$'s efficient AdaLN-Single architecture, reducing parameters while maintaining quality. Experiments on LAION-Aesthetics show that, relative to the training scale reported for prior DDM work, our approach reduces the compute by 16$\times$ and data by 14$\times$. Under aligned inference settings, our heterogeneous configuration achieves better FID and higher intra-prompt diversity than the homogeneous baseline. By eliminating synchronization requirements and enabling mixed DDPM/FM objectives, our framework makes decentralized generative model training accessible to contributors with single GPUs requiring only 24--48GB VRAM.

Cynthia Huang, Yuqing Xie, Zhiying Jiang et al.

In this work, we conceptualize the learning process as information compression. We seek to equip generative pre-trained models with human-like learning capabilities that enable data compression during inference. We present a novel approach that utilizes the Generative Pre-trained Transformer (GPT) to approximate Kolmogorov complexity, with the aim of estimating the optimal Information Distance for few-shot learning. We first propose using GPT as a prior for lossless text compression, achieving a noteworthy compression ratio. Experiment with LLAMA2-7B backbone achieves a compression ratio of 15.5 on enwik9. We justify the pre-training objective of GPT models by demonstrating its equivalence to the compression length, and, consequently, its ability to approximate the information distance for texts. Leveraging the approximated information distance, our method allows the direct application of GPT models in quantitative text similarity measurements. Experiment results show that our method overall achieves superior performance compared to embedding and prompt baselines on challenging NLP tasks, including semantic similarity, zero and one-shot text classification, and zero-shot text ranking.

Zhiying Jiang, Yiqin Dai, Ji Xin et al.

Most real-world problems that machine learning algorithms are expected to solve face the situation with 1) unknown data distribution; 2) little domain-specific knowledge; and 3) datasets with limited annotation. We propose Non-Parametric learning by Compression with Latent Variables (NPC-LV), a learning framework for any dataset with abundant unlabeled data but very few labeled ones. By only training a generative model in an unsupervised way, the framework utilizes the data distribution to build a compressor. Using a compressor-based distance metric derived from Kolmogorov complexity, together with few labeled data, NPC-LV classifies without further training. We show that NPC-LV outperforms supervised methods on all three datasets on image classification in low data regime and even outperform semi-supervised learning methods on CIFAR-10. We demonstrate how and when negative evidence lowerbound (nELBO) can be used as an approximate compressed length for classification. By revealing the correlation between compression rate and classification accuracy, we illustrate that under NPC-LV, the improvement of generative models can enhance downstream classification accuracy.

Jinyuan Liu, Xingyuan Li, Qingyun Mei et al.

Infrared and visible image fusion (IVIF) integrates complementary modalities to enhance scene perception. Current methods predominantly focus on optimizing handcrafted losses and objective metrics, often resulting in fusion outcomes that do not align with human visual preferences. This challenge is further exacerbated by the ill-posed nature of IVIF, which severely limits its effectiveness in human perceptual environments such as security surveillance and driver assistance systems. To address these limitations, we propose a feedback reinforcement framework that bridges human evaluation to infrared and visible image fusion. To address the lack of human-centric evaluation metrics and data, we introduce the first large-scale human feedback dataset for IVIF, containing multidimensional subjective scores and artifact annotations, and enriched by a fine-tuned large language model with expert review. Based on this dataset, we design a domain-specific reward function and train a reward model to quantify perceptual quality. Guided by this reward, we fine-tune the fusion network through Group Relative Policy Optimization, achieving state-of-the-art performance that better aligns fused images with human aesthetics. Code is available at https://github.com/ALKA-Wind/EVAFusion.

Zengxi Zhang, Zhiying Jiang, Jinyuan Liu et al.

Underwater images suffer from light refraction and absorption, which impairs visibility and interferes the subsequent applications. Existing underwater image enhancement methods mainly focus on image quality improvement, ignoring the effect on practice. To balance the visual quality and application, we propose a heuristic normalizing flow for detection-driven underwater image enhancement, dubbed WaterFlow. Specifically, we first develop an invertible mapping to achieve the translation between the degraded image and its clear counterpart. Considering the differentiability and interpretability, we incorporate the heuristic prior into the data-driven mapping procedure, where the ambient light and medium transmission coefficient benefit credible generation. Furthermore, we introduce a detection perception module to transmit the implicit semantic guidance into the enhancement procedure, where the enhanced images hold more detection-favorable features and are able to promote the detection performance. Extensive experiments prove the superiority of our WaterFlow, against state-of-the-art methods quantitatively and qualitatively.

Zhiying Jiang, Zengxi Zhang, Jinyuan Liu et al.

Multi-spectral image stitching leverages the complementarity between infrared and visible images to generate a robust and reliable wide field-of-view (FOV) scene. The primary challenge of this task is to explore the relations between multi-spectral images for aligning and integrating multi-view scenes. Capitalizing on the strengths of Graph Convolutional Networks (GCNs) in modeling feature relationships, we propose a spatial graph reasoning based multi-spectral image stitching method that effectively distills the deformation and integration of multi-spectral images across different viewpoints. To accomplish this, we embed multi-scale complementary features from the same view position into a set of nodes. The correspondence across different views is learned through powerful dense feature embeddings, where both inter- and intra-correlations are developed to exploit cross-view matching and enhance inner feature disparity. By introducing long-range coherence along spatial and channel dimensions, the complementarity of pixel relations and channel interdependencies aids in the reconstruction of aligned multi-view features, generating informative and reliable wide FOV scenes. Moreover, we release a challenging dataset named ChaMS, comprising both real-world and synthetic sets with significant parallax, providing a new option for comprehensive evaluation. Extensive experiments demonstrate that our method surpasses the state-of-the-arts.

Ji Xin, Raphael Tang, Zhiying Jiang et al.

There exists a wide variety of efficiency methods for natural language processing (NLP) tasks, such as pruning, distillation, dynamic inference, quantization, etc. We can consider an efficiency method as an operator applied on a model. Naturally, we may construct a pipeline of multiple efficiency methods, i.e., to apply multiple operators on the model sequentially. In this paper, we study the plausibility of this idea, and more importantly, the commutativity and cumulativeness of efficiency operators. We make two interesting observations: (1) Efficiency operators are commutative -- the order of efficiency methods within the pipeline has little impact on the final results; (2) Efficiency operators are also cumulative -- the final results of combining several efficiency methods can be estimated by combining the results of individual methods. These observations deepen our understanding of efficiency operators and provide useful guidelines for their real-world applications.

Zhiying Jiang, Zengxi Zhang, Jinyuan Liu et al.

Since the differences in viewing range, resolution and relative position, the multi-modality sensing module composed of infrared and visible cameras needs to be registered so as to have more accurate scene perception. In practice, manual calibration-based registration is the most widely used process, and it is regularly calibrated to maintain accuracy, which is time-consuming and labor-intensive. To cope with these problems, we propose a scene-adaptive infrared and visible image registration. Specifically, in regard of the discrepancy between multi-modality images, an invertible translation process is developed to establish a modality-invariant domain, which comprehensively embraces the feature intensity and distribution of both infrared and visible modalities. We employ homography to simulate the deformation between different planes and develop a hierarchical framework to rectify the deformation inferred from the proposed latent representation in a coarse-to-fine manner. For that, the advanced perception ability coupled with the residual estimation conducive to the regression of sparse offsets, and the alternate correlation search facilitates a more accurate correspondence matching. Moreover, we propose the first ground truth available misaligned infrared and visible image dataset, involving three synthetic sets and one real-world set. Extensive experiments validate the effectiveness of the proposed method against the state-of-the-arts, advancing the subsequent applications.

Xingyuan Li, Songcheng Du, Yang Zou et al.

Image fusion aims to integrate complementary information from multiple source images to produce a more informative and visually consistent representation, benefiting both human perception and downstream vision tasks. Despite recent progress, most existing fusion methods are designed for specific tasks (i.e., multi-modal, multi-exposure, or multi-focus fusion) and struggle to effectively preserve source information during the fusion process. This limitation primarily arises from task-specific architectures and the degradation of source information caused by deep-layer propagation. To overcome these issues, we propose UniFusion, a unified image fusion framework designed to achieve cross-task generalization. First, leveraging DINOv3 for modality-consistent feature extraction, UniFusion establishes a shared semantic space for diverse inputs. Second, to preserve the understanding of each source image, we introduce a reconstruction-alignment loss to maintain consistency between fused outputs and inputs. Finally, we employ a bilevel optimization strategy to decouple and jointly optimize reconstruction and fusion objectives, effectively balancing their coupling relationship and ensuring smooth convergence. Extensive experiments across multiple fusion tasks demonstrate UniFusion's superior visual quality, generalization ability, and adaptability to real-world scenarios. Code is available at https://github.com/dusongcheng/UniFusion.

Yang Zou, Xingyue Zhu, Kaiqi Han et al.

Infrared video has been of great interest in visual tasks under challenging environments, but often suffers from severe atmospheric turbulence and compression degradation. Existing video super-resolution (VSR) methods either neglect the inherent modality gap between infrared and visible images or fail to restore turbulence-induced distortions. Directly cascading turbulence mitigation (TM) algorithms with VSR methods leads to error propagation and accumulation due to the decoupled modeling of degradation between turbulence and resolution. We introduce HATIR, a Heat-Aware Diffusion for Turbulent InfraRed Video Super-Resolution, which injects heat-aware deformation priors into the diffusion sampling path to jointly model the inverse process of turbulent degradation and structural detail loss. Specifically, HATIR constructs a Phasor-Guided Flow Estimator, rooted in the physical principle that thermally active regions exhibit consistent phasor responses over time, enabling reliable turbulence-aware flow to guide the reverse diffusion process. To ensure the fidelity of structural recovery under nonuniform distortions, a Turbulence-Aware Decoder is proposed to selectively suppress unstable temporal cues and enhance edge-aware feature aggregation via turbulence gating and structure-aware attention. We built FLIR-IVSR, the first dataset for turbulent infrared VSR, comprising paired LR-HR sequences from a FLIR T1050sc camera (1024 X 768) spanning 640 diverse scenes with varying camera and object motion conditions. This encourages future research in infrared VSR. Project page: https://github.com/JZ0606/HATIR

Ali Rouzbayani, Bidhan Roy, Marcos Villagra et al.

We present Paris 2.0, the first video generation model pre-trained through decentralized computation. Its training recipe builds upon Paris 1.0 (arXiv:2510.03434), the first ever open-weight Decentralized Diffusion Model (DDM), which showed that image generation can be trained without a monolithic GPU cluster. However, temporally coherent video generation had remained an open problem under decentralized training, and Paris 2.0 closes it. In low-resolution text-to-video training, against a monolithic model trained on the same data under a matched total compute budget, Paris 2.0 cuts Frechet Video Distance (FVD) from 561.04 to 279.01, a ~2.0x improvement, and lifts CLIP text-video similarity and aesthetic score.

Xingyuan Li, Haoyuan Xu, Shulin Li et al.

Infrared and visible video fusion is essential for achieving comprehensive perception in dynamic scenes. However, maintaining temporal consistency remains a formidable challenge. Conventional methods relying on optical flow often suffer from geometric rigidity and ghosting artifacts. Moreover, standard diffusion-based fusion models typically operate in a frame-by-frame manner; when extended to autoregressive settings, they lack intrinsic temporal constraints and are prone to severe error accumulation and drifting, where minor artifacts amplify over time. To address these limitations, we propose a drift-resilient video fusion method that reformulates the task as history-conditioned motion generation. We introduce Stabilized History Guidance and Soft Temporal Anchoring to reframe temporal consistency as spectral filtering, implicitly aggregating motion dynamics without rigid alignment. Furthermore, our Decoupled Structure-Motion Adaptation strategy bridges pre-trained priors and structural constraints via two-stage training and latent refinement. Extensive experiments demonstrate that our method achieves state-of-the-art performance in both fusion quality and temporal stability.

Zhiying Jiang, Matthew Y. R. Yang, Mikhail Tsirlin et al.

Deep neural networks (DNNs) are often used for text classification tasks as they usually achieve high levels of accuracy. However, DNNs can be computationally intensive with billions of parameters and large amounts of labeled data, which can make them expensive to use, to optimize and to transfer to out-of-distribution (OOD) cases in practice. In this paper, we propose a non-parametric alternative to DNNs that's easy, light-weight and universal in text classification: a combination of a simple compressor like gzip with a $k$-nearest-neighbor classifier. Without any training, pre-training or fine-tuning, our method achieves results that are competitive with non-pretrained deep learning methods on six in-distributed datasets. It even outperforms BERT on all five OOD datasets, including four low-resource languages. Our method also performs particularly well in few-shot settings where labeled data are too scarce for DNNs to achieve a satisfying accuracy.

Zhiying Jiang, Rui Wang, Dongbo Bu et al.

We aim to bridge the gap between our common-sense few-sample human learning and large-data machine learning. We derive a theory of human-like few-shot learning from von-Neuman-Landauer's principle. modelling human learning is difficult as how people learn varies from one to another. Under commonly accepted definitions, we prove that all human or animal few-shot learning, and major models including Free Energy Principle and Bayesian Program Learning that model such learning, approximate our theory, under Church-Turing thesis. We find that deep generative model like variational autoencoder (VAE) can be used to approximate our theory and perform significantly better than baseline models including deep neural networks, for image recognition, low resource language processing, and character recognition.

Jinyuan Liu, Guanyao Wu, Zhu Liu et al.

Infrared-visible image fusion (IVIF) is a critical task in computer vision, aimed at integrating the unique features of both infrared and visible spectra into a unified representation. Since 2018, the field has entered the deep learning era, with an increasing variety of approaches introducing a range of networks and loss functions to enhance visual performance. However, challenges such as data compatibility, perception accuracy, and efficiency remain. Unfortunately, there is a lack of recent comprehensive surveys that address this rapidly expanding domain. This paper fills that gap by providing a thorough survey covering a broad range of topics. We introduce a multi-dimensional framework to elucidate common learning-based IVIF methods, from visual enhancement strategies to data compatibility and task adaptability. We also present a detailed analysis of these approaches, accompanied by a lookup table clarifying their core ideas. Furthermore, we summarize performance comparisons, both quantitatively and qualitatively, focusing on registration, fusion, and subsequent high-level tasks. Beyond technical analysis, we discuss potential future directions and open issues in this area. For further details, visit our GitHub repository: https://github.com/RollingPlain/IVIF_ZOO.

Jinyuan Liu, Bowei Zhang, Qingyun Mei et al.

Infrared and visible image fusion integrates information from distinct spectral bands to enhance image quality by leveraging the strengths and mitigating the limitations of each modality. Existing approaches typically treat image fusion and subsequent high-level tasks as separate processes, resulting in fused images that offer only marginal gains in task performance and fail to provide constructive feedback for optimizing the fusion process. To overcome these limitations, we propose a Discriminative Cross-Dimension Evolutionary Learning Framework, termed DCEvo, which simultaneously enhances visual quality and perception accuracy. Leveraging the robust search capabilities of Evolutionary Learning, our approach formulates the optimization of dual tasks as a multi-objective problem by employing an Evolutionary Algorithm (EA) to dynamically balance loss function parameters. Inspired by visual neuroscience, we integrate a Discriminative Enhancer (DE) within both the encoder and decoder, enabling the effective learning of complementary features from different modalities. Additionally, our Cross-Dimensional Embedding (CDE) block facilitates mutual enhancement between high-dimensional task features and low-dimensional fusion features, ensuring a cohesive and efficient feature integration process. Experimental results on three benchmarks demonstrate that our method significantly outperforms state-of-the-art approaches, achieving an average improvement of 9.32% in visual quality while also enhancing subsequent high-level tasks. The code is available at https://github.com/Beate-Suy-Zhang/DCEvo.

Zhiying Jiang, Xingyuan Li, Jinyuan Liu et al.

Image stitching seamlessly integrates images captured from varying perspectives into a single wide field-of-view image. Such integration not only broadens the captured scene but also augments holistic perception in computer vision applications. Given a pair of captured images, subtle perturbations and distortions which go unnoticed by the human visual system tend to attack the correspondence matching, impairing the performance of image stitching algorithms. In light of this challenge, this paper presents the first attempt to improve the robustness of image stitching against adversarial attacks. Specifically, we introduce a stitching-oriented attack~(SoA), tailored to amplify the alignment loss within overlapping regions, thereby targeting the feature matching procedure. To establish an attack resistant model, we delve into the robustness of stitching architecture and develop an adaptive adversarial training~(AAT) to balance attack resistance with stitching precision. In this way, we relieve the gap between the routine adversarial training and benign models, ensuring resilience without quality compromise. Comprehensive evaluation across real-world and synthetic datasets validate the deterioration of SoA on stitching performance. Furthermore, AAT emerges as a more robust solution against adversarial perturbations, delivering superior stitching results. Code is available at:https://github.com/Jzy2017/TRIS.

Zengxi Zhang, Zhiying Jiang, Long Ma et al.

Underwater images are often affected by light refraction and absorption, reducing visibility and interfering with subsequent applications. Existing underwater image enhancement methods primarily focus on improving visual quality while overlooking practical implications. To strike a balance between visual quality and application, we propose a heuristic invertible network for underwater perception enhancement, dubbed HUPE, which enhances visual quality and demonstrates flexibility in handling other downstream tasks. Specifically, we introduced an information-preserving reversible transformation with embedded Fourier transform to establish a bidirectional mapping between underwater images and their clear images. Additionally, a heuristic prior is incorporated into the enhancement process to better capture scene information. To further bridge the feature gap between vision-based enhancement images and application-oriented images, a semantic collaborative learning module is applied in the joint optimization process of the visual enhancement task and the downstream task, which guides the proposed enhancement model to extract more task-oriented semantic features while obtaining visually pleasing images. Extensive experiments, both quantitative and qualitative, demonstrate the superiority of our HUPE over state-of-the-art methods. The source code is available at https://github.com/ZengxiZhang/HUPE.

Xingyuan Li, Zirui Wang, Yang Zou et al.

Infrared imaging is essential for autonomous driving and robotic operations as a supportive modality due to its reliable performance in challenging environments. Despite its popularity, the limitations of infrared cameras, such as low spatial resolution and complex degradations, consistently challenge imaging quality and subsequent visual tasks. Hence, infrared image super-resolution (IISR) has been developed to address this challenge. While recent developments in diffusion models have greatly advanced this field, current methods to solve it either ignore the unique modal characteristics of infrared imaging or overlook the machine perception requirements. To bridge these gaps, we propose DifIISR, an infrared image super-resolution diffusion model optimized for visual quality and perceptual performance. Our approach achieves task-based guidance for diffusion by injecting gradients derived from visual and perceptual priors into the noise during the reverse process. Specifically, we introduce an infrared thermal spectrum distribution regulation to preserve visual fidelity, ensuring that the reconstructed infrared images closely align with high-resolution images by matching their frequency components. Subsequently, we incorporate various visual foundational models as the perceptual guidance for downstream visual tasks, infusing generalizable perceptual features beneficial for detection and segmentation. As a result, our approach gains superior visual results while attaining State-Of-The-Art downstream task performance. Code is available at https://github.com/zirui0625/DifIISR

Yang Zou, Jun Ma, Zhidong Jiao et al.

Infrared image super-resolution (IISR) under real-world conditions is a practically significant yet rarely addressed task. Pioneering works are often trained and evaluated on simulated datasets or neglect the intrinsic differences between infrared and visible imaging. In practice, however, real infrared images are affected by coupled optical and sensing degradations that jointly deteriorate both structural sharpness and thermal fidelity. To address these challenges, we propose Real-IISR, a unified autoregressive framework for real-world IISR that progressively reconstructs fine-grained thermal structures and clear backgrounds in a scale-by-scale manner via thermal-structural guided visual autoregression. Specifically, a Thermal-Structural Guidance module encodes thermal priors to mitigate the mismatch between thermal radiation and structural edges. Since non-uniform degradations typically induce quantization bias, Real-IISR adopts a Condition-Adaptive Codebook that dynamically modulates discrete representations based on degradation-aware thermal priors. Also, a Thermal Order Consistency Loss enforces a monotonic relation between temperature and pixel intensity, ensuring relative brightness order rather than absolute values to maintain physical consistency under spatial misalignment and thermal drift. We build FLIR-IISR, a real-world IISR dataset with paired LR-HR infrared images acquired via automated focus variation and motion-induced blur. Extensive experiments demonstrate the promising performance of Real-IISR, providing a unified foundation for real-world IISR and benchmarking. The dataset and code are available at: https://github.com/JZD151/Real-IISR.

Xingyuan Li, Haoyuan Xu, Xingyue Zhu et al.

Infrared and Visible Image Fusion (IVIF) has shown promise in visual tasks under challenging environments, but fusion under unregistered conditions faces inherent misalignments. Current studies to solve them either predict the deformation parameters coarse-to-fine (i.e., coarse registration and fine registration) or estimate the deformation fields in multi-scales for registration. Though straightforward, they overlook the cumulative errors in registration, which contaminate the fusion stage and severely deteriorate the resulting images. We introduce the Spatial-Frequency Registration and Fusion (SFRF) framework, which incorporates uncertainty estimation and infrared thermal radiation distribution consistency into a unified pipeline to handle the error accumulation for robust registration and fusion across both spatial and frequency domains. Specifically, SFRF constructs a Multi-scale Iterative Registration (MIR) framework that iteratively refines the deformation field across scales, leveraging uncertainty estimation at each stage to mitigate error accumulation and enhance alignment accuracy dynamically. To ensure the accurate alignment of infrared thermal distributions during registration, thermal radiation distribution consistency is employed as a frequency-domain supervisory signal, promoting global consistency in the frequency domain. Based on the spatial-frequency alignment, SFRF further adopts a Dual-branch Spatial-Frequency Fusion (DSFF) module, which incorporates spatial geometric features and frequency distribution information to reconstruct visually appealing images. SFRF achieves impressive performance across diverse datasets.

Zhiying Jiang, Ruhao Yan, Zengxi Zhang et al.

Image stitching synthesizes images captured from multiple perspectives into a single image with a broader field of view. The significant variations in object depth often lead to large parallax, resulting in ghosting and misalignment in the stitched results. To address this, we propose a depth-consistency-constrained seamless-free image stitching method. First, to tackle the multi-view alignment difficulties caused by parallax, a multi-stage mechanism combined with global depth regularization constraints is developed to enhance the alignment accuracy of the same apparent target across different depth ranges. Second, during the multi-view image fusion process, an optimal stitching seam is determined through graph-based low-cost computation, and a soft-seam region is diffused to precisely locate transition areas, thereby effectively mitigating alignment errors induced by parallax and achieving natural and seamless stitching results. Furthermore, considering the computational overhead in the shift regression process, a reparameterization strategy is incorporated to optimize the structural design, significantly improving algorithm efficiency while maintaining optimal performance. Extensive experiments demonstrate the superior performance of the proposed method against the existing methods. Code is available at https://github.com/DLUT-YRH/DSFN.

Jinyuan Liu, Zihang Chen, Zhu Liu et al.

We engage in the relatively underexplored task named thermal infrared image enhancement. Existing infrared image enhancement methods primarily focus on tackling individual degradations, such as noise, contrast, and blurring, making it difficult to handle coupled degradations. Meanwhile, all-in-one enhancement methods, commonly applied to RGB sensors, often demonstrate limited effectiveness due to the significant differences in imaging models. In sight of this, we first revisit the imaging mechanism and introduce a Progressive Prompt Fusion Network (PPFN). Specifically, the PPFN initially establishes prompt pairs based on the thermal imaging process. For each type of degradation, we fuse the corresponding prompt pairs to modulate the model's features, providing adaptive guidance that enables the model to better address specific degradations under single or multiple conditions. In addition, a Selective Progressive Training (SPT) mechanism is introduced to gradually refine the model's handling of composite cases to align the enhancement process, which not only allows the model to remove camera noise and retain key structural details, but also enhancing the overall contrast of the thermal image. Furthermore, we introduce the most high-quality, multi-scenarios infrared benchmark covering a wide range of scenarios. Extensive experiments substantiate that our approach not only delivers promising visual results under specific degradation but also significantly improves performance on complex degradation scenes, achieving a notable 8.76\% improvement. Code is available at https://github.com/Zihang-Chen/HM-TIR.

Rodrigo Nogueira, Zhiying Jiang, Jimmy Lin

The ability to perform arithmetic tasks is a remarkable trait of human intelligence and might form a critical component of more complex reasoning tasks. In this work, we investigate if the surface form of a number has any influence on how sequence-to-sequence language models learn simple arithmetic tasks such as addition and subtraction across a wide range of values. We find that how a number is represented in its surface form has a strong influence on the model's accuracy. In particular, the model fails to learn addition of five-digit numbers when using subwords (e.g., "32"), and it struggles to learn with character-level representations (e.g., "3 2"). By introducing position tokens (e.g., "3 10e1 2"), the model learns to accurately add and subtract numbers up to 60 digits. We conclude that modern pretrained language models can easily learn arithmetic from very few examples, as long as we use the proper surface representation. This result bolsters evidence that subword tokenizers and positional encodings are components in current transformer designs that might need improvement. Moreover, we show that regardless of the number of parameters and training examples, models cannot learn addition rules that are independent of the length of the numbers seen during training. Code to reproduce our experiments is available at https://github.com/castorini/transformers-arithmetic

Zhiying Jiang, Raphael Tang, Ji Xin et al.

Pretrained transformers achieve the state of the art across tasks in natural language processing, motivating researchers to investigate their inner mechanisms. One common direction is to understand what features are important for prediction. In this paper, we apply information bottlenecks to analyze the attribution of each feature for prediction on a black-box model. We use BERT as the example and evaluate our approach both quantitatively and qualitatively. We show the effectiveness of our method in terms of attribution and the ability to provide insight into how information flows through layers. We demonstrate that our technique outperforms two competitive methods in degradation tests on four datasets. Code is available at https://github.com/bazingagin/IBA.

Xingyuan Li, Yang Zou, Jinyuan Liu et al.

With the rapid progression of deep learning technologies, multi-modality image fusion has become increasingly prevalent in object detection tasks. Despite its popularity, the inherent disparities in how different sources depict scene content make fusion a challenging problem. Current fusion methodologies identify shared characteristics between the two modalities and integrate them within this shared domain using either iterative optimization or deep learning architectures, which often neglect the intricate semantic relationships between modalities, resulting in a superficial understanding of inter-modal connections and, consequently, suboptimal fusion outcomes. To address this, we introduce a text-guided multi-modality image fusion method that leverages the high-level semantics from textual descriptions to integrate semantics from infrared and visible images. This method capitalizes on the complementary characteristics of diverse modalities, bolstering both the accuracy and robustness of object detection. The codebook is utilized to enhance a streamlined and concise depiction of the fused intra- and inter-domain dynamics, fine-tuned for optimal performance in detection tasks. We present a bilevel optimization strategy that establishes a nexus between the joint problem of fusion and detection, optimizing both processes concurrently. Furthermore, we introduce the first dataset of paired infrared and visible images accompanied by text prompts, paving the way for future research. Extensive experiments on several datasets demonstrate that our method not only produces visually superior fusion results but also achieves a higher detection mAP over existing methods, achieving state-of-the-art results.

Zeru Shi, Zengxi Zhang, Kemeng Cui et al.

Images captured in harsh environments often exhibit blurred details, reduced contrast, and color distortion, which hinder feature detection and matching, thereby affecting the accuracy and robustness of homography estimation. While visual enhancement can improve contrast and clarity, it may introduce visual-tolerant artifacts that obscure the structural integrity of images. Considering the resilience of semantic information against environmental interference, we propose a semantic-driven feature enhancement network for robust homography estimation, dubbed SeFENet. Concretely, we first introduce an innovative hierarchical scale-aware module to expand the receptive field by aggregating multi-scale information, thereby effectively extracting image features under diverse harsh conditions. Subsequently, we propose a semantic-guided constraint module combined with a high-level perceptual framework to achieve degradation-tolerant with semantic feature. A meta-learning-based training strategy is introduced to mitigate the disparity between semantic and structural features. By internal-external alternating optimization, the proposed network achieves implicit semantic-wise feature enhancement, thereby improving the robustness of homography estimation in adverse environments by strengthening the local feature comprehension and context information extraction. Experimental results under both normal and harsh conditions demonstrate that SeFENet significantly outperforms SOTA methods, reducing point match error by at least 41% on the large-scale datasets.

Marcos Villagra, Bidhan Roy, Raihan Seraj et al.

Decentralized Diffusion Models (DDMs) route denoising through experts trained independently on disjoint data clusters, which can strongly disagree in their predictions. What governs the quality of generations in such systems? We present the first ever systematic investigation of this question. A priori, the expectation is that minimizing denoising trajectory sensitivity -- minimizing how perturbations amplify during sampling -- should govern generation quality. We demonstrate this hypothesis is incorrect: a stability-quality dissociation. Full ensemble routing, which combines all expert predictions at each step, achieves the most stable sampling dynamics and best numerical convergence while producing the worst generation quality (FID 47.9 vs. 22.6 for sparse Top-2 routing). Instead, we identify expert-data alignment as the governing principle: generation quality depends on routing inputs to experts whose training distribution covers the current denoising state. Across two distinct DDM systems, we validate expert-data alignment using (i) data-cluster distance analysis, confirming sparse routing selects experts with data clusters closest to the current denoising state, and (ii) per-expert analysis, showing selected experts produce more accurate predictions than non-selected ones, and (iii) expert disagreement analysis, showing quality degrades when experts disagree. For DDM deployment, our findings establish that routing should prioritize expert-data alignment over numerical stability metrics.

Huiming Yang, Wenzhuo Liu, Yicheng Qiao et al.

The sparse cross-modality detector offers more advantages than its counterpart, the Bird's-Eye-View (BEV) detector, particularly in terms of adaptability for downstream tasks and computational cost savings. However, existing sparse detectors overlook the quality of token representation, leaving it with a sub-optimal foreground quality and limited performance. In this paper, we identify that the geometric structure preserved and the class distribution are the key to improving the performance of the sparse detector, and propose a Sparse Selector (SS). The core module of SS is Ray-Aware Supervision (RAS), which preserves rich geometric information during the training stage, and Class-Balanced Supervision, which adaptively reweights the salience of class semantics, ensuring that tokens associated with small objects are retained during token sampling. Thereby, outperforming other sparse multi-modal detectors in the representation of tokens. Additionally, we design Ray Positional Encoding (Ray PE) to address the distribution differences between the LiDAR modality and the image. Finally, we integrate the aforementioned module into an end-to-end sparse multi-modality detector, dubbed CrossRay3D. Experiments show that, on the challenging nuScenes benchmark, CrossRay3D achieves state-of-the-art performance with 72.4 mAP and 74.7 NDS, while running 1.84 faster than other leading methods. Moreover, CrossRay3D demonstrates strong robustness even in scenarios where LiDAR or camera data are partially or entirely missing.

Zhiying Jiang, Raihan Seraj, Marcos Villagra et al.

We present Paris, the first publicly released diffusion model pre-trained entirely through decentralized computation. Paris demonstrates that high-quality text-to-image generation can be achieved without centrally coordinated infrastructure. Paris is open for research and commercial use. Paris required implementing our Distributed Diffusion Training framework from scratch. The model consists of 8 expert diffusion models (129M-605M parameters each) trained in complete isolation with no gradient, parameter, or intermediate activation synchronization. Rather than requiring synchronized gradient updates across thousands of GPUs, we partition data into semantically coherent clusters where each expert independently optimizes its subset while collectively approximating the full distribution. A lightweight transformer router dynamically selects appropriate experts at inference, achieving generation quality comparable to centrally coordinated baselines. Eliminating synchronization enables training on heterogeneous hardware without specialized interconnects. Empirical validation confirms that Paris's decentralized training maintains generation quality while removing the dedicated GPU cluster requirement for large-scale diffusion models. Paris achieves this using 14$\times$ less training data and 16$\times$ less compute than the prior decentralized baseline.

Xiaoke Shang, Gehui Li, Zhiying Jiang et al.

The correction of exposure-related issues is a pivotal component in enhancing the quality of images, offering substantial implications for various computer vision tasks. Historically, most methodologies have predominantly utilized spatial domain recovery, offering limited consideration to the potentialities of the frequency domain. Additionally, there has been a lack of a unified perspective towards low-light enhancement, exposure correction, and multi-exposure fusion, complicating and impeding the optimization of image processing. In response to these challenges, this paper proposes a novel methodology that leverages the frequency domain to improve and unify the handling of exposure correction tasks. Our method introduces Holistic Frequency Attention and Dynamic Frequency Feed-Forward Network, which replace conventional correlation computation in the spatial-domain. They form a foundational building block that facilitates a U-shaped Holistic Dynamic Frequency Transformer as a filter to extract global information and dynamically select important frequency bands for image restoration. Complementing this, we employ a Laplacian pyramid to decompose images into distinct frequency bands, followed by multiple restorers, each tuned to recover specific frequency-band information. The pyramid fusion allows a more detailed and nuanced image restoration process. Ultimately, our structure unifies the three tasks of low-light enhancement, exposure correction, and multi-exposure fusion, enabling comprehensive treatment of all classical exposure errors. Benchmarking on mainstream datasets for these tasks, our proposed method achieves state-of-the-art results, paving the way for more sophisticated and unified solutions in exposure correction.

Zengxi Zhang, Zeru Shi, Zhiying Jiang et al.

Due to the uneven absorption of different light wavelengths in aquatic environments, underwater images suffer from low visibility and clear color deviations. With the advancement of autonomous underwater vehicles, extensive research has been conducted on learning-based underwater enhancement algorithms. These works can generate visually pleasing enhanced images and mitigate the adverse effects of degraded images on subsequent perception tasks. However, learning-based methods are susceptible to the inherent fragility of adversarial attacks, causing significant disruption in enhanced results. In this work, we introduce a collaborative adversarial resilience network, dubbed CARNet, for underwater image enhancement and subsequent detection tasks. Concretely, we first introduce an invertible network with strong perturbation-perceptual abilities to isolate attacks from underwater images, preventing interference with visual quality enhancement and perceptual tasks. Furthermore, an attack pattern discriminator is introduced to adaptively identify and eliminate various types of attacks. Additionally, we propose a bilevel attack optimization strategy to heighten the robustness of the network against different types of attacks under the collaborative adversarial training of vision-driven and perception-driven attacks. Extensive experiments demonstrate that the proposed method outputs visually appealing enhancement images and performs an average 6.71% higher detection mAP than state-of-the-art methods.

Gehui Li, Jinyuan Liu, Long Ma et al.

Photographs taken with less-than-ideal exposure settings often display poor visual quality. Since the correction procedures vary significantly, it is difficult for a single neural network to handle all exposure problems. Moreover, the inherent limitations of convolutions, hinder the models ability to restore faithful color or details on extremely over-/under- exposed regions. To overcome these limitations, we propose a Macro-Micro-Hierarchical transformer, which consists of a macro attention to capture long-range dependencies, a micro attention to extract local features, and a hierarchical structure for coarse-to-fine correction. In specific, the complementary macro-micro attention designs enhance locality while allowing global interactions. The hierarchical structure enables the network to correct exposure errors of different scales layer by layer. Furthermore, we propose a contrast constraint and couple it seamlessly in the loss function, where the corrected image is pulled towards the positive sample and pushed away from the dynamically generated negative samples. Thus the remaining color distortion and loss of detail can be removed. We also extend our method as an image enhancer for low-light face recognition and low-light semantic segmentation. Experiments demonstrate that our approach obtains more attractive results than state-of-the-art methods quantitatively and qualitatively.

Zhiying Jiang, Risheng Liu, Shuzhou Yang et al.

Rain streaks significantly decrease the visibility of captured images and are also a stumbling block that restricts the performance of subsequent computer vision applications. The existing deep learning-based image deraining methods employ manually crafted networks and learn a straightforward projection from rainy images to clear images. In pursuit of better deraining performance, they focus on elaborating a more complicated architecture rather than exploiting the intrinsic properties of the positive and negative information. In this paper, we propose a contrastive learning-based image deraining method that investigates the correlation between rainy and clear images and leverages a contrastive prior to optimize the mutual information of the rainy and restored counterparts. Given the complex and varied real-world rain patterns, we develop a recursive mechanism. It involves multi-scale feature extraction and dynamic cross-level information recruitment modules. The former advances the portrayal of diverse rain patterns more precisely, while the latter can selectively compensate high-level features for shallow-level information. We term the proposed recursive dynamic multi-scale network with a contrastive prior, RDMC. Extensive experiments on synthetic benchmarks and real-world images demonstrate that the proposed RDMC delivers strong performance on the depiction of rain streaks and outperforms the state-of-the-art methods. Moreover, a practical evaluation of object detection and semantic segmentation shows the effectiveness of the proposed method.

Rodrigo Nogueira, Zhiying Jiang, Jimmy Lin

This work proposes a novel adaptation of a pretrained sequence-to-sequence model to the task of document ranking. Our approach is fundamentally different from a commonly-adopted classification-based formulation of ranking, based on encoder-only pretrained transformer architectures such as BERT. We show how a sequence-to-sequence model can be trained to generate relevance labels as "target words", and how the underlying logits of these target words can be interpreted as relevance probabilities for ranking. On the popular MS MARCO passage ranking task, experimental results show that our approach is at least on par with previous classification-based models and can surpass them with larger, more-recent models. On the test collection from the TREC 2004 Robust Track, we demonstrate a zero-shot transfer-based approach that outperforms previous state-of-the-art models requiring in-dataset cross-validation. Furthermore, we find that our approach significantly outperforms an encoder-only model in a data-poor regime (i.e., with few training examples). We investigate this observation further by varying target words to probe the model's use of latent knowledge.

Rodrigo Nogueira, Zhiying Jiang, Kyunghyun Cho et al.

Citation recommendation systems for the scientific literature, to help authors find papers that should be cited, have the potential to speed up discoveries and uncover new routes for scientific exploration. We treat this task as a ranking problem, which we tackle with a two-stage approach: candidate generation followed by re-ranking. Within this framework, we adapt to the scientific domain a proven combination based on "bag of words" retrieval followed by re-scoring with a BERT model. We experimentally show the effects of domain adaptation, both in terms of pretraining on in-domain data and exploiting in-domain vocabulary. In addition, we introduce a novel navigation-based document expansion strategy to enrich the candidate documents processed by our neural models. On three different collections from different scientific disciplines, we achieve the best-reported results in the citation recommendation task.

Qingyun Wang, Lifu Huang, Zhiying Jiang et al.

We present a PaperRobot who performs as an automatic research assistant by (1) conducting deep understanding of a large collection of human-written papers in a target domain and constructing comprehensive background knowledge graphs (KGs); (2) creating new ideas by predicting links from the background KGs, by combining graph attention and contextual text attention; (3) incrementally writing some key elements of a new paper based on memory-attention networks: from the input title along with predicted related entities to generate a paper abstract, from the abstract to generate conclusion and future work, and finally from future work to generate a title for a follow-on paper. Turing Tests, where a biomedical domain expert is asked to compare a system output and a human-authored string, show PaperRobot generated abstracts, conclusion and future work sections, and new titles are chosen over human-written ones up to 30%, 24% and 12% of the time, respectively.

Qingyun Wang, Xiaoman Pan, Lifu Huang et al.

We aim to automatically generate natural language descriptions about an input structured knowledge base (KB). We build our generation framework based on a pointer network which can copy facts from the input KB, and add two attention mechanisms: (i) slot-aware attention to capture the association between a slot type and its corresponding slot value; and (ii) a new \emph{table position self-attention} to capture the inter-dependencies among related slots. For evaluation, besides standard metrics including BLEU, METEOR, and ROUGE, we propose a KB reconstruction based metric by extracting a KB from the generation output and comparing it with the input KB. We also create a new data set which includes 106,216 pairs of structured KBs and their corresponding natural language descriptions for two distinct entity types. Experiments show that our approach significantly outperforms state-of-the-art methods. The reconstructed KB achieves 68.8% - 72.6% F-score.

Risheng Liu, Xin Fan, Minjun Hou et al.

Single image dehazing is an important low-level vision task with many applications. Early researches have investigated different kinds of visual priors to address this problem. However, they may fail when their assumptions are not valid on specific images. Recent deep networks also achieve relatively good performance in this task. But unfortunately, due to the disappreciation of rich physical rules in hazes, large amounts of data are required for their training. More importantly, they may still fail when there exist completely different haze distributions in testing images. By considering the collaborations of these two perspectives, this paper designs a novel residual architecture to aggregate both prior (i.e., domain knowledge) and data (i.e., haze distribution) information to propagate transmissions for scene radiance estimation. We further present a variational energy based perspective to investigate the intrinsic propagation behavior of our aggregated deep model. In this way, we actually bridge the gap between prior driven models and data driven networks and leverage advantages but avoid limitations of previous dehazing approaches. A lightweight learning framework is proposed to train our propagation network. Finally, by introducing a taskaware image separation formulation with a flexible optimization scheme, we extend the proposed model for more challenging vision tasks, such as underwater image enhancement and single image rain removal. Experiments on both synthetic and realworld images demonstrate the effectiveness and efficiency of the proposed framework.