In this paper we tackle a fundamental question: "Can we train latent diffusion models together with the variational auto-encoder (VAE) tokenizer in an end-to-end manner?" Traditional deep-learning wisdom dictates that end-to-end training is often preferable when possible. However, for latent diffusion transformers, it is observed that end-to-end training both VAE and diffusion-model using standard diffusion-loss is ineffective, even causing a degradation in final performance. We show that while diffusion loss is ineffective, end-to-end training can be unlocked through the representation-alignment (REPA) loss -- allowing both VAE and diffusion model to be jointly tuned during the training process. Despite its simplicity, the proposed training recipe (REPA-E) shows remarkable performance; speeding up diffusion model training by over 17x and 45x over REPA and vanilla training recipes, respectively. Interestingly, we observe that end-to-end tuning with REPA-E also improves the VAE itself; leading to improved latent space structure and downstream generation performance. In terms of final performance, our approach sets a new state-of-the-art; achieving FID of 1.26 and 1.83 with and without classifier-free guidance on ImageNet 256x256. Code is available at https://end2end-diffusion.github.io.
论文检索
输入标题、作者或关键词，从 9,256 篇学术成果中精准定位
Few-shot image generation aims to generate diverse and high-quality images for an unseen class given only a few examples in that class. A key challenge in this task is balancing category consistency and image diversity, which often compete with each other. Moreover, existing methods offer limited control over the attributes of newly generated images. In this work, we propose Hyperbolic Diffusion Autoencoders (HypDAE), a novel approach that operates in hyperbolic space to capture hierarchical relationships among images from seen categories. By leveraging pre-trained foundation models, HypDAE generates diverse new images for unseen categories with exceptional quality by varying stochastic subcodes or semantic codes. Most importantly, the hyperbolic representation introduces an additional degree of control over semantic diversity through the adjustment of radii within the hyperbolic disk. Extensive experiments and visualizations demonstrate that HypDAE significantly outperforms prior methods by achieving a better balance between preserving category-relevant features and promoting image diversity with limited data. Furthermore, HypDAE offers a highly controllable and interpretable generation process.
Modeling 3D human-object interaction (HOI) is a problem of great interest for computer vision and a key enabler for virtual and mixed-reality applications. Existing methods work in a one-way direction: some recover plausible human interactions conditioned on a 3D object; others recover the object pose conditioned on a human pose. Instead, we provide the first unified model - TriDi which works in any direction. Concretely, we generate Human, Object, and Interaction modalities simultaneously with a new three-way diffusion process, allowing to model seven distributions with one network. We implement TriDi as a transformer attending to the various modalities' tokens, thereby discovering conditional relations between them. The user can control the interaction either as a text description of HOI or a contact map. We embed these two representations onto a shared latent space, combining the practicality of text descriptions with the expressiveness of contact maps. Using a single network, TriDi unifies all the special cases of prior work and extends to new ones modeling a family of seven distributions. Remarkably, despite using a single model, TriDi generated samples surpass one-way specialized baselines on GRAB and BEHAVE in terms of both qualitative and quantitative metrics, and demonstrating better diversity. We show applicability of TriDi to scene population, generating object for human-contact datasets, and generalization to unseen object geometry.
We propose MikuDance, a diffusion-based pipeline incorporating mixed motion dynamics to animate stylized character art. MikuDance consists of two key techniques: Mixed Motion Modeling and Mixed-Control Diffusion, to address the challenges of high-dynamic motion and reference-guidance misalignment in character art animation. Specifically, a Scene Motion Tracking strategy is presented to explicitly model the dynamic camera in pixel-wise space, enabling unified character-scene motion modeling. Building on this, the Mixed-Control Diffusion implicitly aligns the scale and body shape of diverse characters with motion guidance, allowing flexible control of local character motion. Subsequently, a Motion-Adaptive Normalization module is incorporated to effectively inject global scene motion, paving the way for comprehensive character art animation. Through extensive experiments, we demonstrate the effectiveness and generalizability of MikuDance across various character art and motion guidance, consistently producing high-quality animations with remarkable motion dynamics.
Typical template-based object pose pipelines estimate the pose by retrieving the closest matching template and aligning it with the observed image. However, failure to retrieve the correct template often leads to inaccurate pose predictions. To address this, we reformulate template-based object pose estimation as a ray alignment problem, where the viewing directions from multiple posed template images are learned to align with a non-posed query image. Inspired by recent progress in diffusion-based camera pose estimation, we embed this formulation into a diffusion transformer architecture that aligns a query image with a set of posed templates. We reparameterize object rotation using object-centered camera rays and model object translation by extending scale-invariant translation estimation to dense translation offsets. Our model leverages geometric priors from the templates to guide accurate query pose inference. A coarse-to-fine training strategy based on narrowed template sampling improves performance without modifying the network architecture. Extensive experiments across multiple benchmark datasets show competitive results of our method compared to state-of-the-art approaches in unseen object pose estimation.
We present MagicMirror, a framework for generating identity-preserved videos with cinematic-level quality and dynamic motion. While recent advances in video diffusion models have shown impressive capabilities in text-to-video generation, maintaining consistent identity while producing natural motion remains challenging. Previous methods either require person-specific fine-tuning or struggle to balance identity preservation with motion diversity. Built upon Video Diffusion Transformers, our method introduces three key components: (1) a dual-branch facial feature extractor that captures both identity and structural features, (2) a lightweight cross-modal adapter with Conditioned Adaptive Normalization for efficient identity integration, and (3) a two-stage training strategy combining synthetic identity pairs with video data. Extensive experiments demonstrate that MagicMirror effectively balances identity consistency with natural motion, outperforming existing methods across multiple metrics while requiring minimal parameters added. The code and model will be made publicly available.
Reinforcement learning (RL) has proven its potential in complex decision-making tasks. Yet, many RL systems rely on manually crafted state representations, requiring effort in feature engineering. Visual Reinforcement Learning (VRL) offers a way to address this challenge by enabling agents to learn directly from raw visual input. Nonetheless, VRL continues to face generalization issues, as models often overfit to specific domain features.To tackle this issue, we propose Diffusion Guided Adaptive Augmentation (DGA2), an augmentation method that utilizes Stable Diffusion to enhance domain diversity.We introduce an Adaptive Domain Shift strategy that dynamically adjusts the degree of domain shift according to the agent's learning progress for effective augmentation with Stable Diffusion.Additionally, we employ saliency as the mask to preserve the semantics of data.Our experiments on the DMControl-GB, Adroit, and Procgen environments demonstrate that DGA2 improves generalization performance compared to existing data augmentation and generalization methods.
Decoding Correlation-Induced Misalignment in the Stable Diffusion Workflow for Text-to-Image Generation
PDF ↗The fundamental requirement for text-to-image generation is aligning the generated images with the provided text. With large-scale data, pre-trained Stable Diffusion (SD) models have achieved remarkable performance in this task. These models process an input prompt as text control, guiding a vision model to perform denoising operations that recover a clean image from pure noise. However, we observe that when there is correlation among text tokens, SD's generated images fail to accurately represent the semantics of the input prompt: simple yet crucial objects may be omitted, thereby disrupting text-image alignment. We refer to this problem as "object omission". Without additional external knowledge, previous methods have been ineffective at addressing this issue. To investigate this problem, we analyze the attention maps in SD and find that biased text representations mislead the visual denoising process when handling correlated tokens, impeding object generation. Moreover, we observe that even when two prompts share the same semantics, slight variations in token sequence significantly alter attention scores, consequently affecting the final generated images. Based on these findings, we propose a simple yet effective fine-tuning method that applies decorrelation to the self-attention maps in the text module, thus reducing dependencies between tokens. Our approach requires no external prior knowledge, is straightforward to implement, and operates solely on the text module of the SD model. Extensive experiments confirm that our method effectively alleviates the object omission problem under text correlations, thereby enhancing text-image alignment.
Multimodal Diffusion Transformers (MM-DiTs) have recently emerged as a powerful framework for unified text-vision synthesis, surpassing traditional U-Net architectures in generative tasks. One key innovation lies in its Multimodal Self-Attention (MM-SA) interaction where image and text tokens are concatenated and processed via self-attention.However, this mechanism poses significant challenges for editing, rendering conventional U-Net-based attention manipulation methods ineffective. To address this limitation, we propose QK-Edit, a training-free framework that exploits the unique attention dynamics of MM-DiTs for precise text-guided image and video editing. By introducing a novel query-key manipulation strategy, our method isolates and adjusts critical attention components to achieve an optimal balance between prompt fidelity and structural consistency. This enables seamless edits across various tasks, including object addition, object removal, object replacement, changing background, changing material, changing color, and style transformation. Notably, it can be easily implemented with feature replacement in inference.QK-Edit demonstrates superior editing performance on state-of-the-art models, such as FLUX and HunyuanVideo, effectively bridging the gap between generative power and editable flexibility in MM-DiTs, and paving the way for scalable multimodal content creation.
Diffusion models have achieved remarkable success in image generation but come with significant computational costs, posing challenges for deployment in resource-constrained environments. Recent post-training quantization (PTQ) methods have attempted to mitigate this issue by focusing on the iterative nature of diffusion models. However, these approaches often overlook outliers, leading to degraded performance at low bit-widths. In this paper, we propose a DMQ which combines Learned Equivalent Scaling (LES) and channel-wise Power-of-Two Scaling (PTS) to effectively address these challenges. Learned Equivalent Scaling optimizes channel-wise scaling factors to redistribute quantization difficulty between weights and activations, reducing overall quantization error. Recognizing that early denoising steps, despite having small quantization errors, crucially impact the final output due to error accumulation, we incorporate an adaptive timestep weighting scheme to prioritize these critical steps during learning. Furthermore, identifying that layers such as skip connections exhibit high inter-channel variance, we introduce channel-wise Power-of-Two Scaling for activations. To ensure robust selection of PTS factors even with small calibration set, we introduce a voting algorithm that enhances reliability. Extensive experiments demonstrate that our method significantly outperforms existing works, especially at low bit-widths such as W4A6 (4-bit weight, 6-bit activation) and W4A8, maintaining high image generation quality and model stability. The code is available at https://github.com/LeeDongYeun/dmq.
Diffusion models have emerged as a powerful technique for text-to-image (T2I) generation, creating high-quality, diverse images across various domains. However, a common limitation in these models is the incomplete display of objects, where fragments or missing parts can undermine the model's performance in downstream applications such as dataset synthesis and video generation using 2D prior-based models. % that demand visual accuracy, such as e-commerce product imaging and realistic digital content creation.In this study, we conduct the in-depth analysis of this issue and reveal that the primary culprit behind incomplete object generation is RandomCrop. This data augmentation method, widely used in diffusion models, though enhances model generalization ability, disrupts object continuity during training. To address this, we propose a training-free solution that penalizes activation values occurring at image boundaries during the early denoising steps. Our method is easily applicable to pre-trained Stable Diffusion models with minimal modifications and negligible computational overhead. Extensive experiments demonstrate the effectiveness of our method, showing substantial improvements in object integrity and image quality.
D3QE: Learning Discrete Distribution Discrepancy-aware Quantization Error for Autoregressive-Generated Image Detection
PDF ↗The emergence of visual autoregressive (AR) models has revolutionized image generation while presenting new challenges for synthetic image detection. Unlike previous GAN or diffusion-based methods, AR models generate images through discrete token prediction, exhibiting both marked improvements in image synthesis quality and unique characteristics in their vector-quantized representations. In this paper, we propose to leverage Discrete Distribution Discrepancy-aware Quantization Error (D^3QE) for autoregressive-generated image detection that exploits the distinctive patterns and the frequency distribution bias of the codebook existing in real and fake images. We introduce a discrete distribution discrepancy-aware transformer that integrates dynamic codebook frequency statistics into its attention mechanism, fusing semantic features and quantization error latent. To evaluate our method, we construct a comprehensive dataset termed ARForensics covering 7 mainstream visual AR models. Experiments demonstrate superior detection accuracy and strong generalization of D^3QE across different AR models, with robustness to real-world perturbations. Code is available at \href https://github.com/Zhangyr2022/D3QE https://github.com/Zhangyr2022/D3QE .
Photolithography Overlay Map Generation with Implicit Knowledge Distillation Diffusion Transformer
PDF ↗This paper presents the Implicit Knowledge Distillation Diffusion Transformer (IKDDiT), a groundbreaking model tailored for photolithography overlay map generation in semiconductor manufacturing. IKDDiT effectively addresses the challenges of open-vocabulary overlay map generation by integrating pre-trained image-text encoders, diffusion models, and masked transformers. Utilizing advanced text-to-image diffusion and image-text discriminative models, it generates high-fidelity overlay maps across multiple photolithography layers, significantly mitigating overlay misregistration errors and minimizing productivity losses caused by wafer rework. Key innovations include an implicit knowledge distillation framework that refines inter-image alignment by decoupling discriminative and generative tasks via an implicit discriminator, as well as a gated cross-attention mechanism to enhance generative performance. Experimental results demonstrate that IKDDiT achieves an optimal trade-off between efficiency and accuracy, providing a scalable, robust solution poised to advance overlay map generation in semiconductor processes.
While multi-step diffusion models have advanced both forward and inverse rendering, existing approaches often treat these problems independently, leading to cycle inconsistency and slow inference speed. In this work, we present Ouroboros, a framework composed of two single-step diffusion models that handle forward and inverse rendering with mutual reinforcement. Our approach extends intrinsic decomposition to both indoor and outdoor scenes and introduces a cycle consistency mechanism that ensures coherence between forward and inverse rendering outputs. Experimental results demonstrate state-of-the-art performance across diverse scenes while achieving substantially faster inference speed compared to other diffusion-based methods. We also demonstrate that Ouroboros can transfer to video decomposition in a training-free manner, reducing temporal inconsistency in video sequences while maintaining high-quality per-frame inverse rendering.
Lens flare significantly degrades image quality, impacting critical computer vision tasks like object detection and autonomous driving. Recent Single Image Flare Removal (SIFR) methods perform poorly when off-frame light sources are incomplete or absent. We propose LightsOut, a diffusion-based outpainting framework tailored to enhance SIFR by reconstructing off-frame light sources. Our method leverages a multitask regression module and LoRA fine-tuned diffusion model to ensure realistic and physically consistent outpainting results. Comprehensive experiments demonstrate LightsOut consistently boosts the performance of existing SIFR methods across challenging scenarios without additional retraining, serving as a universally applicable plug-and-play preprocessing solution.
FoundIR: Unleashing Million-scale Training Data to Advance Foundation Models for Image Restoration
PDF ↗Despite the significant progress made by all-in-one models in universal image restoration, existing methods suffer from a generalization bottleneck in real-world scenarios, as they are mostly trained on small-scale synthetic datasets with limited degradations. Therefore, large-scale high-quality real-world training data is urgently needed to facilitate the emergence of foundation models for image restoration. To advance this field, we spare no effort in contributing a million-scale dataset with two notable advantages over existing training data: larger-scale real-world samples, and higher-diversity data types. By adjusting internal camera settings and external imaging conditions, we can capture aligned image pairs using our well-designed data acquisition system over multiple rounds and our data alignment criterion. Moreover, we propose a robust model, FoundIR, to better address a broader range of restoration tasks in real-world scenarios, taking a further step toward foundation models. Specifically, we first utilize a diffusion-based generalist model to remove degradations by learning the degradation-agnostic common representations from diverse inputs, where incremental learning strategy is adopted to better guide model training. To refine the model's restoration capability in complex scenarios, we introduce degradation-aware specialist models for achieving final high-quality results. Extensive experiments show the value of our dataset and the effectiveness of our method.
Text-guided image and 3D editing have advanced with diffusion-based models, yet methods like Delta Denoising Score often struggle with stability, spatial control, and editing strength. These limitations stem from reliance on complex auxiliary structures, which introduce conflicting optimization signals and restrict precise, localized edits. We introduce Stable Score Distillation (SSD), a streamlined framework that enhances stability and alignment in the editing process by anchoring a single classifier to the source prompt. Specifically, SSD utilizes Classifier-Free Guidance (CFG) equation to achieve cross-prompt alignment, and introduces a constant term null-text branch to stabilize the optimization process. This approach preserves the original content's structure and ensures that editing trajectories are closely aligned with the source prompt, enabling smooth, prompt-specific modifications while maintaining coherence in surrounding regions. Additionally, SSD incorporates a prompt enhancement branch to boost editing strength, particularly for style transformations. Our method achieves state-of-the-art results in 2D and 3D editing tasks, including NeRF and text-driven style edits, with faster convergence and reduced complexity, providing a robust and efficient solution for text-guided editing. Code is available at: https://github.com/Alex-Zhu1/SSD.
Most 3D object generators prioritize aesthetic quality, often neglecting the physical constraints necessary for practical applications. One such constraint is that a 3D object should be self-supporting, i.e., remain balanced under gravity. Previous approaches to generating stable 3D objects relied on differentiable physics simulators to optimize geometry at test time, which is slow, unstable, and prone to local optima. Inspired by the literature on aligning generative models with external feedback, we propose Direct Simulation Optimization (DSO). This framework leverages feedback from a (non-differentiable) simulator to increase the likelihood that the 3D generator directly outputs stable 3D objects. We construct a dataset of 3D objects labeled with stability scores obtained from the physics simulator. This dataset enables fine-tuning of the 3D generator using the stability score as an alignment metric, via direct preference optimization (DPO) or direct reward optimization (DRO)---a novel objective we introduce to align diffusion models without requiring pairwise preferences. Our experiments demonstrate that the fine-tuned feed-forward generator, using either the DPO or DRO objective, is significantly faster and more likely to produce stable objects than test-time optimization. Notably, the DSO framework functions even without any ground-truth 3D objects for training, allowing the 3D generator to self-improve by automatically collecting simulation feedback on its own outputs.
3D shape generation has greatly flourished through the development of so-called "native" 3D diffusion, particularly through the Vectset Diffusion Model (VDM). While recent advancements have shown promising results in generating high-resolution 3D shapes, VDM still struggles at high-speed generation. Challenges exist because of not only difficulties in accelerating diffusion sampling but also VAE decoding in VDM -- areas under-explored in previous works. To address these challenges, we present FlashVDM, a systematic framework for accelerating both VAE and DiT in VDM. For DiT, FlashVDM enables flexible diffusion sampling with as few as 5 inference steps, while maintaining comparable quality, which is made possible by stabilizing consistency distillation with our newly introduced Progressive Flow Distillation technique. For VAE, we introduce a lightning vectset decoder equipped with Adaptive KV Selection, Hierarchical Volume Decoding,, and Efficient Network Design. By exploiting the locality of vectset and the sparsity of shape surface in the volume, the proposed decoder drastically lowers FLOPs, minimizing the overall decoding overhead. We apply FlashVDM to the current state-of-the-art open-source shape generation model Hunyuan3D-2, resulting in Hunyuan3D-2 Turbo. Through systematic evaluation for both generation and reconstruction, we demonstrate that our model outperforms existing fast 3D generation methods by a significant margin, achieving comparable performance to the state-of-the-art models while reducing inference time by over 45x for reconstruction and 32x for generation. Code and models are publicly available at https://github.com/Tencent-Hunyuan/FlashVDM.
Text-to-image (T2I) diffusion models excel at generating photorealistic images, but commonly struggle to render accurate spatial relationships described in text prompts. We identify two core issues underlying this common failure: 1) the ambiguous nature of spatial-related data in existing datasets, and 2) the inability of current text encoders to accurately interpret the spatial semantics of input descriptions. We address these issues with CoMPaSS, a versatile training framework that enhances spatial understanding of any T2I diffusion model. CoMPaSS solves the ambiguity of spatial-related data with the Spatial Constraints-Oriented Pairing (SCOP) data engine, which curates spatially accurate training data through a set of principled spatial constraints. To better exploit the curated high-quality spatial priors, CoMPaSS further introduces a Token ENcoding ORdering (TENOR) module to allow better exploitation of high-quality spatial priors, effectively compensating for the shortcoming of text encoders. Extensive experiments on four popular open-weight T2I diffusion models covering both UNet- and MMDiT-based architectures demonstrate the effectiveness of CoMPaSS by setting new state-of-the-arts with substantial relative gains across well-known benchmarks on spatial relationships generation, including VISOR (+98%), T2I-CompBench Spatial (+67%), and GenEval Position (+131%).