arXiv subjects

Ge Wu

Publications and source records attributed to Ge Wu.

17 recordsLinked to original sources

arXiv2026

Exploring the Performance Frontier of Compact Unified Image Generation Models

We present Swift-Image, a compact unified model for text-to-image generation, single-image editing, and multi-image editing. Our goal is to explore how far a relatively small visual generator can be pushed through systematic training engineering under a constrained computational budget. Swift-Image adopts an efficient 6B single-stream DiT and a progressive training pipeline that evolves from broad semantic coverage to higher resolution, stronger visual quality, and unified generation-editing supervision. For post-training, we employ parallel expert reinforcement learning followed by multi-teacher on-policy distillation to alleviate interference among heterogeneous objectives. We further decouple high-level reasoning from pixel-level rendering with a Prompt Enhancer that translates user requests into generator-aligned visual specifications. For efficient deployment, structural pruning and few-step distillation produce 3B and accelerated variants. Swift-Image achieves leading aggregate performance among evaluated open-source models with only 6B parameters and 243K GPU training hours; the compressed 3B model incurs nearly no loss, while few-step distillation further improves aggregate editing performance with substantially fewer sampling steps. Our study also summarizes practical lessons for architecture, data curriculum, post-training, prompt enhancement, and model compression.

cs.CV↗

arXiv2026

Structural Leakage in Graph Encryption: Attacks and Defenses

Graph encryption schemes (GES) enable secure outsourcing of graph data while supporting efficient queries. This report provides a comprehensive analysis of structural leakage in GES for single-pair shortest path (SPSP) queries, integrating findings from two recent works. First, we analyze PathGES, a scheme designed to resist query recovery attacks through heavy-light decomposition (HLD) and canonical fragment encoding. Our analysis reveals that PathGES suffers from significant imbalances in HLD decomposition, with over 99% of token-path mappings being one-to-one on real-world datasets, enabling both the Falzon-Paterson attack and side-channel inference of path lengths. Second, we present Fragment Tree attack that exploits these structural weaknesses to recover query contents, achieving up to 10.24% exact recovery on sparse graphs. Third, we introduce BlindGES, an enhanced scheme incorporating a Merge-and-Divide mechanism and two-level multimap index that reduces one-to-one mappings to below 20%, cuts setup time by 50%, reduces storage overhead by 32%, and limits path length leakage to under 1%. This report systematically presents attack methodologies, defense mechanisms, security proofs, and experimental evaluations on seven real-world datasets.

cs.CR↗

arXiv2026

GraphPO: Graph-based Policy Optimization for Reasoning Models

Reinforcement Learning with Verifiable Rewards (RLVR) has become a standard paradigm for enhancing the capability of large reasoning models. RLVR typically samples responses independently and optimizes the policy using from final answers. This paradigm has two limitations. First, independently responses often contain similar intermediate reasoning steps, causing redundant exploration and wasted computation. Second, sparse final-answer rewards make it hard to identify useful steps. Tree-based methods partly address this problem by sharing prefixes and comparing branches from the same prefix to provide fine-grained signals. However, tree branches are still expanded independently. When different branches reach similar reasoning states, they cannot share information and repeat similar exploration. Moreover, tree-based methods ignore such dispersion and only perform local comparisons within separate branches, which can lead to higher variance in advantage estimation. To address this challenge, we propose GraphPO (Graph-based Policy Optimization), a novel RL framework that represents rollouts as a directed acyclic graph, with reasoning steps as edges and semantic states summarized from the reasoning paths as nodes. GraphPO merges semantically equivalent reasoning paths into equivalence classes, allowing them to share suffixes and reallocating budget away from redundant expansions to diverse exploration. Furthermore, we assign efficiency advantages to incoming edges and correctness advantages to outgoing edges, thereby improving inference efficiency while deriving process supervision from outcome. Theory shows that GraphPO reduces advantage-estimation variance and enhances reasoning efficiency. Experiments on three LLMs across reasoning and agentic search benchmarks show that GraphPO consistently outperforms chain- and tree-based baselines with the same token budgets or response budgets.

cs.CL↗

arXiv2026

RefAlign: Representation Alignment for Reference-to-Video Generation

Reference-to-video (R2V) generation is a controllable video synthesis paradigm that constrains the generation process using both text prompts and reference images, enabling applications such as personalized advertising and virtual try-on. In practice, existing R2V methods typically introduce additional high-level semantic or cross-modal features alongside the VAE latent representation of the reference image and jointly feed them into the diffusion Transformer (DiT). These auxiliary representations provide semantic guidance and act as implicit alignment signals, which can partially alleviate pixel-level information leakage in the VAE latent space. However, they may still struggle to address copy--paste artifacts and multi-subject confusion caused by modality mismatch across heterogeneous encoder features. In this paper, we propose RefAlign, a representation alignment framework that explicitly aligns DiT reference-branch features to the semantic space of a visual foundation model (VFM). The core of RefAlign is a reference alignment loss that pulls the reference features and VFM features of the same subject closer to improve identity consistency, while pushing apart the corresponding features of different subjects to enhance semantic discriminability. This simple yet effective strategy is applied only during training, incurring no inference-time overhead, and achieves a better balance between text controllability and reference fidelity. Extensive experiments on the OpenS2V-Eval benchmark demonstrate that RefAlign outperforms current state-of-the-art methods in TotalScore, validating the effectiveness of explicit reference alignment for R2V tasks.

cs.CV↗

arXiv2026

WaDi: Weight Direction-aware Distillation for One-step Image Synthesis

Despite the impressive performance of diffusion models such as Stable Diffusion (SD) in image generation, their slow inference limits practical deployment. Recent works accelerate inference by distilling multi-step diffusion into one-step generators. To better understand the distillation mechanism, we analyze U-Net/DiT weight changes between one-step students and their multi-step teacher counterparts. Our analysis reveals that changes in weight direction significantly exceed those in weight norm, highlighting it as the key factor during distillation. Motivated by this insight, we propose the Low-rank Rotation of weight Direction (LoRaD), a parameter-efficient adapter tailored to one-step diffusion distillation. LoRaD is designed to model these structured directional changes using learnable low-rank rotation matrices. We further integrate LoRaD into Variational Score Distillation (VSD), resulting in Weight Direction-aware Distillation (WaDi)-a novel one-step distillation framework. WaDi achieves state-of-the-art FID scores on COCO 2014 and COCO 2017 while using only approximately 10% of the trainable parameters of the U-Net/DiT. Furthermore, the distilled one-step model demonstrates strong versatility and scalability, generalizing well to various downstream tasks such as controllable generation, relation inversion, and high-resolution synthesis.

cs.CV↗

arXiv2026

Microstructural origin of the simultaneous enhancements in strength and ductility of a nitrogen-doped high-entropy alloy

As one of the most abundant interstitial elements, nitrogen (N) is effective in improving yield strength of metallic materials, due to interstitial solid solution strengthening. Doping N can substantially enhance the yield strength but often leads to a decreased ductility, revealing a strength-ductility trade-off phenomenon. Here, we simultaneously enhance the strength and ductility in a non-equiatomic CrMnFeCoNi high-entropy alloy via N alloying and unravel the underlying microscopic mechanisms. The N-doped alloy (1 at.% N) shows an excellent combination of higher yield strength (104% increase) and larger ductility (38% increase), with a two-stage strain hardening behavior, compared to the N-free alloy. Detailed transmission electron microscopy (TEM) analysis reveals that N-doping introduces short-range order (SRO) domains within the microstructure, leads to pronounced planar slip, and promotes the formation of nano-spaced (6-15 nm) stacking faults and deformation twins. Continuous generation and interaction of the fine-spaced SFs act as a strong barrier for dislocation movement and provide ample room for dislocation storage. The interaction of SRO with dislocations and the evolution of SFs ascribe to the first strain hardening stage, and the disordering of the SRO along with the activation of deformation twins are attributed to the second strain hardening stage. Our work shows that N-doping is effective in simultaneously improving the strength-ductility synergy and provides novel insights into alloy design with slightly elevating the SFE, and manipulating the ordered structure within the HEA.

cond-mat.mtrl-sci↗

arXiv2025

MeanFlow Transformers with Representation Autoencoders

MeanFlow (MF) is a diffusion-motivated generative model that enables efficient few-step generation by learning long jumps directly from noise to data. In practice, it is often used as a latent MF by leveraging the pre-trained Stable Diffusion variational autoencoder (SD-VAE) for high-dimensional data modeling. However, MF training remains computationally demanding and is often unstable. During inference, the SD-VAE decoder dominates the generation cost, and MF depends on complex guidance hyperparameters for class-conditional generation. In this work, we develop an efficient training and sampling scheme for MF in the latent space of a Representation Autoencoder (RAE), where a pre-trained vision encoder (e.g., DINO) provides semantically rich latents paired with a lightweight decoder. We observe that naive MF training in the RAE latent space suffers from severe gradient explosion. To stabilize and accelerate training, we adopt Consistency Mid-Training for trajectory-aware initialization and use a two-stage scheme: distillation from a pre-trained flow matching teacher to speed convergence and reduce variance, followed by an optional bootstrapping stage with a one-point velocity estimator to further reduce deviation from the oracle mean flow. This design removes the need for guidance, simplifies training configurations, and reduces computation in both training and sampling. Empirically, our method achieves a 1-step FID of 2.03, outperforming vanilla MF's 3.43, while reducing sampling GFLOPS by 38% and total training cost by 83% on ImageNet 256. We further scale our approach to ImageNet 512, achieving a competitive 1-step FID of 3.23 with the lowest GFLOPS among all baselines. Code is available at https://github.com/sony/mf-rae.

cs.CV↗

arXiv2025

Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance

Current image super-resolution methods show strong performance on natural images but distort text, creating a fundamental trade-off between image quality and textual readability. To address this, we introduce TIGER (Text-Image Guided supEr-Resolution), a novel two-stage framework that breaks this trade-off through a "text-first, image-later" paradigm. TIGER explicitly decouples glyph restoration from image enhancement: it first reconstructs precise text structures and uses them to guide full-image super-resolution. This ensures high fidelity and readability. To support comprehensive training and evaluation, we present the UZ-ST (UltraZoom-Scene Text) dataset, the first Chinese scene text dataset with extreme zoom. Extensive experiments show TIGER achieves state-of-the-art performance, enhancing readability and image quality.

cs.CV↗

arXiv2025

Revisiting Data Challenges of Computational Pathology: A Pack-based Multiple Instance Learning Training Framework

Computational pathology (CPath) digitizes pathology slides into whole slide images (WSIs), enabling analysis for critical healthcare tasks such as cancer diagnosis and prognosis. However, WSIs possess extremely long sequence lengths (up to 200K), significant length variations (from 200 to 200K), and limited supervision. These extreme variations in sequence length lead to high data heterogeneity and redundancy. Conventional methods often compromise on training efficiency and optimization to preserve such heterogeneity under limited supervision. To comprehensively address these challenges, we propose a pack-based MIL framework. It packs multiple sampled, variable-length feature sequences into fixed-length ones, enabling batched training while preserving data heterogeneity. Moreover, we introduce a residual branch that composes discarded features from multiple slides into a hyperslide which is trained with tailored labels. It offers multi-slide supervision while mitigating feature loss from sampling. Meanwhile, an attention-driven downsampler is introduced to compress features in both branches to reduce redundancy. By alleviating these challenges, our approach achieves an accuracy improvement of up to 8% while using only 12% of the training time in the PANDA(UNI). Extensive experiments demonstrate that focusing data challenges in CPath holds significant potential in the era of foundation models. The code is https://github.com/FangHeng/PackMIL

cs.CV↗

arXiv2025

Position: Reasoning After Perception Means Reasoning Without Vision

A common belief in multimodal research is that the perceptual weaknesses of vision--language models can be compensated by stronger language reasoning (e.g., chain-of-thought, in-context learning, or external tools). We challenge this assumption. We argue that for a broad class of visual tasks hard to specify in language, failures stem from a structural fatality where the temporal decision of \textit{when} to reason strictly dictates the spatial constraint of \textit{where} reasoning takes place. When visual reasoning is deferred to language generation, current architectures do not merely delay computation; they displace it from the continuous visual representation to a discrete textual space. Consequently, the sequential ``Perception-then-Reasoning'' paradigm degenerates perception into a passive, one-off feature encoding process, rendering it functionally equivalent to ``Reasoning-in-Text-Space'', where task-critical spatial signals are collapsed before reasoning begins. We substantiate this claim with the Turing Eye Test (TET): tasks that must be resolved in \emph{visual space} and are hard to verbalize; results show text-only reasoning cannot remedy these perceptual failures. Our findings suggest rethinking the architectural divide: shifting from reasoning \textit{about} perception to reasoning \textit{within} perception. This facilitates actively reasoning-driven perception that operates directly on pixel-level visual representations, rather than within a collapsed textual space.

cs.CV↗

arXiv2025

Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think

REPA and its variants effectively mitigate training challenges in diffusion models by incorporating external visual representations from pretrained models, through alignment between the noisy hidden projections of denoising networks and foundational clean image representations. We argue that the external alignment, which is absent during the entire denoising inference process, falls short of fully harnessing the potential of discriminative representations. In this work, we propose a straightforward method called Representation Entanglement for Generation (REG), which entangles low-level image latents with a single high-level class token from pretrained foundation models for denoising. REG acquires the capability to produce coherent image-class pairs directly from pure noise, substantially improving both generation quality and training efficiency. This is accomplished with negligible additional inference overhead, requiring only one single additional token for denoising (<0.5\% increase in FLOPs and latency). The inference process concurrently reconstructs both image latents and their corresponding global semantics, where the acquired semantic knowledge actively guides and enhances the image generation process. On ImageNet 256$\times$256, SiT-XL/2 + REG demonstrates remarkable convergence acceleration, achieving $\textbf{63}\times$ and $\textbf{23}\times$ faster training than SiT-XL/2 and SiT-XL/2 + REPA, respectively. More impressively, SiT-L/2 + REG trained for merely 400K iterations outperforms SiT-XL/2 + REPA trained for 4M iterations ($\textbf{10}\times$ longer). Code is available at: https://github.com/Martinser/REG.

cs.CV↗

arXiv2025

LEDiT: Your Length-Extrapolatable Diffusion Transformer without Positional Encoding

Diffusion transformers (DiTs) struggle to generate images at resolutions higher than their training resolutions. The primary obstacle is that the explicit positional encodings(PE), such as RoPE, need extrapolating to unseen positions which degrades performance when the inference resolution differs from training. In this paper, We propose a Length-Extrapolatable Diffusion Transformer~(LEDiT) to overcome this limitation. LEDiT needs no explicit PEs, thereby avoiding PE extrapolation. The key innovation of LEDiT lies in the use of causal attention. We demonstrate that causal attention can implicitly encode global positional information and show that such information facilitates extrapolation. We further introduce a locality enhancement module, which captures fine-grained local information to complement the global coarse-grained position information encoded by causal attention. Experimental results on both conditional and text-to-image generation tasks demonstrate that LEDiT supports up to 4x resolution scaling (e.g., from 256x256 to 512x512), achieving better image quality compared to the state-of-the-art length extrapolation methods. We believe that LEDiT marks a departure from the standard RoPE-based methods and offers a promising insight into length extrapolation. Project page: https://shenzhang2145.github.io/ledit/

cs.CV↗

arXiv2024

Cascade Prompt Learning for Vision-Language Model Adaptation

Prompt learning has surfaced as an effective approach to enhance the performance of Vision-Language Models (VLMs) like CLIP when applied to downstream tasks. However, current learnable prompt tokens are primarily used for the single phase of adapting to tasks (i.e., adapting prompt), easily leading to overfitting risks. In this work, we propose a novel Cascade Prompt Learning CasPL framework to enable prompt learning to serve both generic and specific expertise (i.e., boosting and adapting prompt) simultaneously. Specifically, CasPL is a new learning paradigm comprising two distinct phases of learnable prompts: the first boosting prompt is crafted to extract domain-general knowledge from a senior larger CLIP teacher model by aligning their predicted logits using extensive unlabeled domain images. The second adapting prompt is then cascaded with the frozen first set to fine-tune the downstream tasks, following the approaches employed in prior research. In this manner, CasPL can effectively capture both domain-general and task-specific representations into explicitly different gradual groups of prompts, thus potentially alleviating overfitting issues in the target domain. It's worth noting that CasPL serves as a plug-and-play module that can seamlessly integrate into any existing prompt learning approach. CasPL achieves a significantly better balance between performance and inference speed, which is especially beneficial for deploying smaller VLM models in resource-constrained environments. Compared to the previous state-of-the-art method PromptSRC, CasPL shows an average improvement of 1.85% for base classes, 3.44% for novel classes, and 2.72% for the harmonic mean over 11 image classification datasets. Code is publicly available at: https://github.com/megvii-research/CasPL.

cs.CV↗

arXiv2023

Blockchain-based Privacy-Preserving Public Key Searchable Encryption with Strong Traceability

Public key searchable encryption (PKSE) scheme allows data users to search over encrypted data. To identify illegal users, many traceable PKSE schemes have been proposed. However, existing schemes cannot trace the keywords which illegal users searched and protect users' privacy simultaneously. In some practical applications, tracing both illegal users' identities and the keywords which they searched is quite important to against the abuse of data. It is a challenge to bind users' identities and keywords while protecting their privacy. Moreover, existing traceable PKSE schemes do not consider the unforgeability and immutability of trapdoor query records, which can lead to the occurrence of frame-up and denying. In this paper, to solve these problems, we propose a blockchain-based privacy-preserving PKSE with strong traceability (BP3KSEST) scheme. Our scheme provides the following features: (1) authorized users can authenticate to trapdoor generation center and obtain trapdoors without releasing their identities and keywords; (2) when data users misbehave in the system, the trusted third party (TTP) can trace both their identities and the keywords which they searched; (3) trapdoor query records are unforgeable; (4) trapdoor query records are immutable because records are stored in blockchain. Notably, this scheme is suitable to the scenarios where privacy must be considered, e.g., electronic health record (EHR). We formalize both the definition and security model of our BP3KSEST scheme, and present a concrete construction. Furthermore, the security of the proposed scheme is formally proven. Finally, the implementation and evaluation are conducted to analyze its efficiency.

cs.CR↗

arXiv2023

A Survey of Historical Learning: Learning Models with Learning History

New knowledge originates from the old. The various types of elements, deposited in the training history, are a large amount of wealth for improving learning deep models. In this survey, we comprehensively review and summarize the topic--``Historical Learning: Learning Models with Learning History'', which learns better neural models with the help of their learning history during its optimization, from three detailed aspects: Historical Type (what), Functional Part (where) and Storage Form (how). To our best knowledge, it is the first survey that systematically studies the methodologies which make use of various historical statistics when training deep neural networks. The discussions with related topics like recurrent/memory networks, ensemble learning, and reinforcement learning are demonstrated. We also expose future challenges of this topic and encourage the community to pay attention to the think of historical learning principles when designing algorithms. The paper list related to historical learning is available at \url{https://github.com/Martinser/Awesome-Historical-Learning.}

cs.LG↗

arXiv2022

Linear Array Network for Low-light Image Enhancement

Convolution neural networks (CNNs) based methods have dominated the low-light image enhancement tasks due to their outstanding performance. However, the convolution operation is based on a local sliding window mechanism, which is difficult to construct the long-range dependencies of the feature maps. Meanwhile, the self-attention based global relationship aggregation methods have been widely used in computer vision, but these methods are difficult to handle high-resolution images because of the high computational complexity. To solve this problem, this paper proposes a Linear Array Self-attention (LASA) mechanism, which uses only two 2-D feature encodings to construct 3-D global weights and then refines feature maps generated by convolution layers. Based on LASA, Linear Array Network (LAN) is proposed, which is superior to the existing state-of-the-art (SOTA) methods in both RGB and RAW based low-light enhancement tasks with a smaller amount of parameters. The code is released in https://github.com/cuiziteng/LASA_enhancement.

cs.CV↗

arXiv2020

AIM 2020 Challenge on Rendering Realistic Bokeh

This paper reviews the second AIM realistic bokeh effect rendering challenge and provides the description of the proposed solutions and results. The participating teams were solving a real-world bokeh simulation problem, where the goal was to learn a realistic shallow focus technique using a large-scale EBB! bokeh dataset consisting of 5K shallow / wide depth-of-field image pairs captured using the Canon 7D DSLR camera. The participants had to render bokeh effect based on only one single frame without any additional data from other cameras or sensors. The target metric used in this challenge combined the runtime and the perceptual quality of the solutions measured in the user study. To ensure the efficiency of the submitted models, we measured their runtime on standard desktop CPUs as well as were running the models on smartphone GPUs. The proposed solutions significantly improved the baseline results, defining the state-of-the-art for practical bokeh effect rendering problem.

eess.IV↗