Latent generative models typically follow a two-stage pipeline, training a variational autoencoder for reconstruction and then a generative model on the frozen latent space. Since reconstruction-optimized latents are not necessarily generation-friendly, jointly training both models is an appealing alternative. However, direct end-to-end training remains challenging, as it is prone to latent collapse and faces a generation-reconstruction conflict. We revisit this problem by analyzing how different objectives shape the latent space and identify two key insights. First, the entropy term in the Kullback-Leibler divergence objective is essential for preventing collapse: reconstruction and prior fitting tend to shrink the posterior, while entropy preserves non-degenerate latent uncertainty. Second, reconstruction and generation exhibit asymmetric learning dynamics: reconstruction is fast and strongly supervised, whereas generation is slower and harder to optimize. Based on these insights, we achieve the first direct end-to-end training without latent collapse and propose GenFirst, a simple generation-before-reconstruction strategy. The generative objective first shapes the latent space under weak reconstruction pressure, after which reconstruction is progressively strengthened to recover visual details. We validate GenFirst with continuous autoregressive priors with exact likelihoods and SiT priors with implicit likelihoods. With our end-to-end objective and GenFirst, SiT achieves a gFID of 0.97 with CFG and 1.45 without CFG on ImageNet-256, while MMDiT reaches a GenEval score of 0.90 on text-to-image generation. Beyond image generation, we extend the framework to shared visual latents for generation and representation learning, and to continuous unified text-image generation. These results demonstrate the generality of stable end-to-end latent learning across generative priors and modalities.
Traditional image similarity metrics such as Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and the Structural Similarity Index Measure (SSIM) rely on pixel-level comparisons and often fail to capture perceptually meaningful differences between images. In contrast, latent representations learned by deep neural networks encode high-level semantic information that is more closely aligned with human visual perception. This paper proposes a convolutional autoencoder-based framework for quantifying image differences using cosine similarity in latent space. The learned compact embeddings enable robust differentiation between visually distinct images under variations in illumination, pose, and background. Extensive evaluation on dog-cat images and additional cross-domain datasets demonstrates clear class-wise clustering and strong inter-class separability in the latent space, with 98.4% of dog-cat image pairs exhibiting similarity scores below 0.5. Further validation using the TID2013 dataset shows that latent-space distance correlates positively with human Mean Opinion Scores (MOS), demonstrating sensitivity to perceptually relevant image distortions. The proposed approach provides a computationally efficient and semantically grounded alternative to conventional pixel-based similarity metrics, with potential applications in content-based retrieval, perceptual quality assessment, and semantic similarity analysis.
Leonardo Zini, Elia Frigieri, Lorenzo Baraldics.CV cs.AI
Scalable Vector Graphics are a fundamental medium for resolution-independent visual content, yet the deep learning community lacks a continuous, dense, and invertible latent space for vector representations, the kind of foundational building block that Variational Autoencoders and their descendants have long provided for raster images. We introduce SLS (SVG Latent Space), a Transformer-based autoencoder that learns compact dense representations of individual SVG paths, the atomic visual elements from which any SVG image can be composed. By modeling SVG commands, coordinate data, and visual properties within a unified BPE-based token vocabulary, SLS learns fixed-size latent representations that jointly capture structure and appearance, and can be decoded back into valid, style-consistent SVG paths with high fidelity. The resulting embedding space is robust, invertible, and structured: embeddings lie on a unit hypersphere, enabling efficient similarity search, composition, and downstream conditioning through simple vector-space operations. Finally, we demonstrate that SLS generalizes across diverse tasks reducing their FLOPs by over 150 times compared to token-based approaches, and establishing a general-purpose latent foundation for vector graphics research.
Fine-grained control over continuous semantic attributes of 3D objects is essential for 3D content creation, but is not well supported by conventional 3D modeling workflows or prompt-based interaction with existing generative AI tools. While slider-based methods have proven effective for fine-grained semantic control in 2D image generation, no equivalent approach exists for 3D. Extending these 2D methods to 3D is non-trivial due to challenges unique to 3D, including geometric integrity and cross-view coherence. We present SemanticSlider3D, a technique for continuous semantic attribute editing of 3D objects that requires no per-attribute training. Given a user-specified attribute, our pipeline constructs a semantic editing direction in the latent space of a state-of-the-art 3D generation model, presenting a diverse and coherent spectrum of 3D variations. A technical validation on a dataset of 50 3D object-attribute pairs shows our method was preferred by all five human assessors across variation range, consistency, 3D object quality, and attribute disentanglement, over a baseline combining a 2D slider with an image-to-3D model. An exploratory study with six participants demonstrates that SemanticSlider3D supported decision-making in 3D prototyping and was perceived as a valuable addition to existing workflows.
Many scalable latent 3D generators operate on structured tensors, whereas pre-optimized 3D Gaussian Splatting (3DGS) reconstructions are unordered, spatially irregular, and vary widely in primitive count. We present GS-Voxel, a fitting-free structured latent framework, and evaluate it for large-scale aerial 3D Gaussian scene generation. GS-Voxel deterministically converts a compatible pre-optimized 3DGS reconstruction into sparse active voxels without additional per-scene optimization, retaining the sub-voxel positions and rendering attributes of the selected primitives. A GS-specific factorized VAE then separately encodes voxel geometry and local Gaussian attributes into sparse 3D latents whose size grows with the number of occupied voxels rather than being limited by a fixed scene-wide primitive count. We train image-conditioned flow models in the GS-Voxel latent space to generate aerial 3DGS scenes. A key application enabled by GS-Voxel is large-area scene generation: overlap-aware tiled inference extends synthesis beyond a single training crop conditioned on satellite-view images. Our results show that GS-Voxel provides structured latents for pre-optimized aerial 3DGS reconstructions, with latent capacity that grows with the number of occupied voxels.
Latent video generation relies on autoencoders to define a compact space in which generative models operate. Although video autoencoder architectures have evolved substantially, their latent spaces are still optimized primarily for pixel-level reconstruction and provide limited high-level semantic organization. A reconstruction-optimal latent space, however, need not be well suited to generative modeling. We propose V-RAE, a video representation autoencoder that builds compact generative latents on top of frozen vision foundation model representations. A lightweight temporal pooling module removes temporal redundancy while preserving semantic structure, and a video decoder reconstructs continuous motion from the compressed features. We evaluate V-RAE with four representative frozen encoders on video reconstruction, semantic probing, and class-conditional generation. V-RAE achieves 2.13 rFVD on K600, outperforming all evaluated large-scale pretrained video VAEs. Its latents retain substantially more semantic information than conventional video tokenizer latents. Under matched generation settings, our best variant achieves gFVD scores of 117.86 and 19.16 on UCF101 and K600, respectively, while converging up to 6x faster}. We further show that reconstruction quality alone is insufficient to characterize generative utility and introduce tFVD, a temporal-coherence diagnostic that correlates more reliably with downstream generation quality. Beyond video generation, V-RAE also improves future video prediction on Cityscapes over the Wan 2.2 VAE latent space under matched prediction settings. Taken together, the experiments show that frozen semantic representations can support video reconstruction, generation, and predictive modeling. The project page: https://v-rae.github.io/.
4D generation synthesizes dynamic 3D scenes from conditions such as text or images. Existing methods either reconstruct generated RGB videos with a separate 4D model or adapt a particular video generator to predict geometry directly. The former suffers from distribution mismatch and error propagation, whereas the latter ties 4D prediction to a specific generator and may require retraining when the generator or conditioning regime changes. We ask whether the final denoised latents of video models that share a variational autoencoder (VAE) can instead provide a reusable interface to explicit 4D prediction. Building on this insight, we introduce direct latent-to-4D generation and instantiate it as Latent-to-4D, which bypasses RGB by aligning a video latent with the token grid of a pretrained 4D decoder and refining it through frame-wise and global spatiotemporal attention. Trained on roughly 1K existing reconstruction clips, a single checkpoint transfers unchanged across multiple video diffusion transformers within the same VAE family. On Text4D-200 and I4D-200, Latent-to-4D surpasses matched same-latent Wan+4RC cascades in projection-based DINO-F1 by 2.88--3.45 and 5.81 points, respectively, while also being preferred by human raters for geometry, temporal stability, and overall quality.
Emilien Seiler, Nicolas Talabot, Yingxuan You +2cs.CV
Optimizing 3D shapes within the latent spaces of deep generative models is fundamental to computer assisted engineering, yet remains prone to a critical failure mode we term manifold drift: the tendency of gradient-based optimization to move latent vectors away from the manifold of valid shapes. This problem is exacerbated in state-of-the-art 3D shape generative models that operate in increasingly high-dimensional latent spaces where valid shapes occupy a vanishingly small fraction of the full space. Existing mitigation strategies, including latent regularization and flow-matching approaches, either sacrifice expressiveness, demand a difficult trade-off between objective guidance and generative fidelity that remains prone to manifold drift, or are computationally infeasible to scale to modern, large-capacity 3D shape models. We introduce a novel optimizer-corrector framework that alternates between gradient steps for objective minimization and guided flow matching to drive the latent state back to the valid shape manifold. By decoupling objective minimization from flow-based correction, optimizing freely and correcting strictly, this alternating design avoids inherent trade-offs, preserving geometric validity without sacrificing expressiveness while remaining computationally feasible on modern 3D shape models. We demonstrate its effectiveness across generative priors of varying complexity, from simple vector latent spaces to large-scale architectures across a variety of downstream optimization tasks, including aerodynamic drag reduction and object compliance optimization.
Direct spectral editing in video-VAE latents can control noise, flicker, smoothness, and frequency content without a decode--filter--reencode pass. However, video VAEs may redistribute pixel-space frequency bands across latent channels, and latent edits can disrupt VAE round-trip dynamics. We introduce \emph{latent-frequency validity} (LFV), which learns a compact VAE-specific spectral response and deploys it only when it improves decoded-target fidelity without worsening round-trip drift. LFV follows a validation-selected path from a diagonal per-frequency calibrator (C1) to full channel mixing (CM), making cross-channel capacity a controllable per-edit resource. Across 544 VAE--edit cells spanning six spectral families, LFV emits 423 cheap operators: 277 are handled by C1, while 146 (34.5\% of emitted operators) require channel mixing. On the primary 120-cell radial sweep, 99/100 emitted operators pass source-video-grouped held-out evaluation. Across five additional filter families, all 323 emitted operators pass held-out evaluation. Fully frozen OpenVid-fitted operators, including the validation-selected path coefficient, pass all 20 tested CogVideoX and HunyuanVideo generated-domain cells without adaptation. The selected response matches direct latent-filter latency and is about $3\times$ faster than pixel filter--reencode. The resulting maps reveal distinct VAE regimes, including strongly channel-coupled CogVideoX responses and a sharp Open-Sora high-band stability frontier.
Forecasting future states from video sequences is a critical challenge for autonomous robotic systems and a fundamental objective of world modeling. Prior generative methods operating at the pixel level inevitably overemphasize task-irrelevant details, leading to prohibitive computational overhead. While latent-based approaches attempt to mitigate this by predicting features directly, the persistent reliance on heavy decoders for state-to-task mapping remains a computational bottleneck. In this work, we propose Decoder-Free Feature Forecasting (DF$^3$), a novel framework that models world evolution entirely within the latent space and directly derives task outputs, completely eliminating the need for a decoder. Specifically, DF$^3$ injects learnable spatial queries into the terminal blocks of a frozen vision foundation model to extract future state representations directly. By employing a lightweight, unified Motion-Aware Context Fusion (MACF) mechanism that seamlessly integrates coarse flow warping with fine-grained latent cross-correlation, these queries interact with historical token representations to explicitly align and forecast the feature of the next frame. Subsequently, a specialized set of task queries probes these forecasted features for the downstream task. Extensive experiments on public benchmarks and zero-shot deployment in a robotic simulator demonstrate that DF$^3$ achieves performance comparable to state-of-the-art methods while offering superior efficiency and flexibility for integrated perception and control.
Latents from vision foundation models (VFMs) are semantically rich and well suited for visual understanding. Recent representation autoencoder methods such as RAE have shown that they can provide promising latent spaces for image generation. However, VFM latents remain difficult to model directly: DiT-generated latents exhibit spectral mismatch with encoder latents, especially in high-frequency components. Our channel-wise spectral analysis further reveals that these high-frequency components are diffusely distributed across latent channels and entangled with semantic information, making the latent space difficult for DiT to model. To address these challenges, we propose SPAE, latent adaptation framework for generation. Specifically, SPAE employs a compact bottleneck to distill stable semantic information while suppressing high-frequency components, thereby improving the alignment between DiT-generated latents and encoder latents. In addition, we apply a channel-wise masking strategy to promote the decoupling of semantic information and high-frequency details across bottleneck channels. Experiments show that SPAE achieves a favorable balance among visual understanding, generation quality, and reconstruction fidelity.
Lisa Weijler, Irene Ballester, Guofeng Mei +2cs.CV
Geometric foundation models, such as the Visual Geometry Grounded Transformer (VGGT), provide strong 3D priors from unposed images. However, such models operate purely in a feed-forward, deterministic regime, \ie~they cannot generate plausible geometry beyond what the input views directly support. Generative models for 3D scenes, on the other hand, must rely on strong geometric priors to produce coherent outputs from sparse inputs. We bridge these two paradigms by performing flow matching directly in VGGT's latent space, leveraging its learned 3D priors without committing to any explicit downstream representation such as Gaussians, meshes, or video-VAE latents. This requires respecting the latent geometry: VGGT tokens occupy a product of high-dimensional hyperspheres on which standard Euclidean flow matching fails. We address this with a Riemannian Flow Matching framework defined on a product manifold of four hyperspheres, aligned with VGGT's multi-scale encoder, which keeps generated tokens on the valid data manifold required by the frozen decoding heads. On RealEstate10K, ScanNet++ and ETH3D, our method achieves strong performance against recent scene generation baselines in both per-view appearance and aggregated 3D geometry, establishing latent-space flow matching on geometric foundation models as a viable paradigm for 3D generation. The project page can be found $\href{https://lisaweijler.github.io/geometry-grounded-rfm/}{\text{here}}$.
Vaddi Charan Sai Nandan Reddy, Harini B, Chandana M Scs.CV cs.LG
We present a novel system that integrates negative prompt optimization via a fine-tuned sequence-to-sequence LLM and latent-space classifier guidance to improve the quality of images generated by Stable Diffusion. Our approach automatically generates optimized negative prompts, and employs a CNN-RNN hybrid classifier to evaluate and guide diffusion steps, rolling back low-quality latent updates. Experimental results demonstrate that our dual-guidance framework reduces artifacts and improves semantic fidelity compared to baseline diffusion.
In this paper, we ask whether vision foundation models construct representations that reflect the intrinsic properties of 3D Euclidean space. Unlike previous works that probe 3D awareness of vision features by regressing image-centric quantities such as depth or normals, we investigate the relation between the structure of the space of visual features and the group of Euclidean transformations $SE(3)$. We propose a set of probes to evaluate this relation from both topological and geometric perspectives: a mutual neighborhood metric that measures the alignment between feature neighborhoods and spatial topology, and a Poincaré Adapter to test the linear accessibility of the geometry of camera motion from latent displacements in static scenes. We show that self-supervised vision models, which, in principle, have not been trained with direct 3D supervision or active agency, possess latent subspaces that are remarkably strongly correlated with three-dimensional Euclidean space, when probed correctly. Building on this insight we propose a new class of "Latent-Space Navigation" techniques that perform visual odometry and localization purely in the latent space, bypassing the need for explicit 3D reconstruction.
Video generative models commonly rely on latent spaces learned by 3D Variational Autoencoders (3D-VAEs). However, conventional 3D-VAEs are mainly optimized for pixel-level reconstruction, which can limit the semantic and spatio-temporal structure captured by their latents. Meanwhile, Video Foundation Models (VFMs) such as V-JEPA 2 and VideoMAEv2 show strong video understanding capabilities, yet whether their frozen representations can be transformed into compact, reconstruction-capable, and generation-friendly video latents remains largely unexplored. We answer this question with VideoRAE, a representation autoencoder that leverages multi-scale hierarchical features from a frozen video foundation encoder and compresses them with a lightweight 1D self-attention projector. VideoRAE supports both continuous latents for Diffusion Transformers and discrete tokens for autoregressive models via multi-codebook high-dimensional quantization. During decoding, a local-and-global representation alignment objective with the frozen VFM teacher improves semantic preservation and enables training without KL regularization. Experiments show that VideoRAE achieves strong reconstruction in both continuous and discrete regimes. On UCF-101, it obtains state-of-the-art class-to-video gFVDs of 40 and 93 with AR and DiT generators, respectively, while converging approximately 5x faster than competing autoencoder baselines. In a controlled 2B-scale text-to-video study, replacing LTX-VAE with VideoRAE leads to faster convergence under comparable settings. These results validate frozen VFM representations as versatile and generation-friendly video latents. The model and code will be released on https://zhxie0117.github.io/VideoRAE.
Daniel Garibi, Ronen Kamenetsky, Hadar Averbuch-Elor +2cs.CV cs.GR
Generating and editing a person's face demands high precision, as even minor modifications can significantly alter a subject's perceived identity. Current personalization and editing methods built on general-purpose text-to-image models, however, often lack the precision required for fine-grained facial edits. We present a method for fine-grained identity tuning in text-to-image personalization models. Unlike standard image editing, which operates on a given image, identity tuning modifies the latent representation of a specific identity, enabling the generation of diverse images that consistently depict the same edited identity. To enable fine-grained latent identity tuning, we explore the latent space of a pre-trained, frozen encoder for text-to-image personalization. Our approach requires no additional training. Instead, it leverages the existing architecture of a frozen encoder to uncover latent semantic directions. This space consists of a set of latent tokens that play distinct roles in capturing different aspects of an identity and often correspond to specific spatial or semantic facial regions. We show that meaningful directions can be identified within this space and within subspaces defined by selected tokens, enabling localized, fine-grained, and semantically coherent edits. We validate our approach through qualitative and quantitative experiments that demonstrate diverse localized facial edits while preserving cross-image identity consistency. Project page at: https://garibida.github.io/IdentityTuning/
Quantum diffusion models provide a physics-consistent route to generative learning by formulating noising and denoising directly on quantum states. However, applying such models to classical high-dimensional data is constrained by the qubit cost of state encoding and the computational burden of simulating large density operators. We propose a scalable hybrid generative pipeline that combines a classical autoencoder for dimensionality reduction with a mixed-state quantum denoising diffusion probabilistic model (MSQuDDPM) operating in the learned latent space. The autoencoder compresses data into compact latent codes that can be embedded into a small-qubit Hilbert space, after which the quantum diffusion model learns a generative distribution over latent density operators and decodes samples back to the original domain. Algorithmically, we simplify the reverse dynamics by predicting an estimate of the clean state $ρ_0$ at timestep $t$ and computing the one-step reverse update via an analytic backward propagation rule, rather than learning an explicit predictor for $ρ_{t-1}$. We demonstrate the proposed approach on MNIST image generation and discuss how mixed-state quantum diffusion can serve as a practical backbone for hybrid quantum--classical generative modeling under realistic qubit budgets.
Diffusion-based generative models have transformed visual content synthesis, yet they remain vulnerable to unauthorized usage and lack reliable attribution methods. Existing watermarking techniques often treat latent tensors as static spatial feature maps or depend on pixel-domain modification, and most do not explicitly leverage the internal frequency structure of the latent space for dual-band redundant embedding, leaving them susceptible to the stochastic nature of diffusion and regeneration attacks. We introduce BiSLW, a trainable bi-spectral latent watermarking framework that jointly embeds aligned identity signals across complementary spectral bands of the decoded diffusion latent using learned encoders and decoders, going beyond fixed-pattern frequency approaches. We leverage the inherent frequency structure of diffusion latents to design a dual-band watermarking framework. Low-frequency components encode global semantics, while high-frequency components capture fine texture. We exploit this structure to embed watermarks across complementary spectral bands. The watermark is independently injected into both bands via learned encoders and recombined before decoding, ensuring it becomes intrinsic to the generative trajectory. Dual spectral decoders recover the watermark from each band, while a cross-band consistency constraint enforces alignment between semantic and textural embeddings. Experiments show that BiSLW achieves a strong balance between perceptual fidelity and robustness, improving PSNR by over 3 dB compared to prior latent diffusion watermarking methods while preserving near-perfect bit accuracy under aggressive regeneration and common distortions, all with negligible computational overhead.
Andrea Sanchietti, Riccardo Marin, Bharat Lal Bhatnagar +2cs.CV
While garments are essential for realistic digital humans, their topological variety makes them much harder to model than parametric bodies. Traditional tailoring relies on 2D sewing patterns, yet bridging these patterns to 3D geometry currently requires physical simulations. We present Stitched Embeddings, the first simulation-free framework to unify 3D garment reconstruction and sewing pattern inference within a single bidirectional latent space. By leveraging the geometric priors of a pretrained 3D foundation model, our approach overcomes the data scarcity typically associated with high-quality garment modeling. We propose to use the BoxMesh as a critical intermediate representation to align 2D panels into 3D configurations without the computational overhead of a simulator. This architecture achieves state-of-the-art accuracy in pattern reconstruction while significantly improving efficiency. Furthermore, our differentiable pipeline enables novel applications, including pattern recovery from meshes and 3D editing from 2D patterns. Finally, this work provides a scalable link between neural 3D vision and the physical garment manufacturing pipeline. Project Page: https://andreus00.github.io/stitchedembeddings
Modern one-step diffusion models achieve impressive quality through distribution-based timestep distillation. Yet, they rely on a critical assumption: Teacher and Student must inhabit the same latent space. This Shared-Space constraint prevents knowledge transfer from modern high-capacity Teachers (e.g., SD 3.5 and Flux) into compact, deployment-friendly Students such as SD 1.5, whose latent resolution and VAE parameterization differ from the Teacher. We formalize this overlooked regime as Cross-Space Distillation, where Teacher and Student differ in both latent resolution and VAE space. To enable distillation under this mismatch, we introduce the Bridge, a lightweight latent interface that maps Student latents into the Teacher space without modifying the Student backbone. Bridge combines a frozen Student VAE decoder as a spatial prior with a compact learnable projector, and is trained with latent reconstruction and attention fidelity objectives for stable Teacher-space alignment. Across diverse modern Teachers, Bridge enables substantial gains for compact one-step Students; for example, it improves SD 1.5 from 5.4 to 9.4 HPSv3 while preserving one-step inference, low latency, and broad ecosystem compatibility. These results show that heterogeneous large Teachers can be distilled into efficient, deployable backbones through a lightweight latent-space interface.
Single-image reflection removal (SIRR) seeks to recover the transmission layer from a mixture corrupted by reflections -- a severely ill-posed problem. Existing methods operate in pixel space, where the nonlinear sRGB formation model entangles the two layers and limits generalization. We observe that pretrained VAE latent spaces exhibit substantially lower coherence between image layers compared to pixel space, providing a more favorable working space for decomposition. Building on this finding, we propose \textbf{PRISM} (Pretrained-latent Reflection Image Separation Model), which reinterprets SIRR as a latent linear separation problem. Under an approximate additive formulation in latent space, PRISM learns a flow matching velocity field on a pretrained FLUX backbone that recovers both transmission and reflection in a single forward pass. To enforce robust disentanglement, we introduce a Latent Composition Consistency (LCC) strategy that constructs synthetic mixtures by swapping reflection latents across samples and enforces consistent decomposition via a cycle loss. We further propose a Layer Contrastive Separation (LCS) loss that promotes semantic separation between layers through patch-level contrastive learning, without requiring explicit reflection targets. Experiments on six benchmarks demonstrate that PRISM consistently outperforms state-of-the-art methods by significant margins, with strong generalization to in-the-wild images.
Class-incremental learning requires a model to learn new classes while preserving decision regions for old ones. This is difficult when raw old samples are no longer available. We propose Prototype Latent World Model Replay, a memory-free framework that stores old classes as distributions over stable hidden states rather than as images. A frozen ImageNet-pretrained encoder maps each image into a latent state space. In this space, each class is summarized by several prototype-centered distributions with class-specific variances. When new classes arrive, the model samples old latent states from this prototype world model. It then trains a lightweight adapter and classifier using both sampled old states and real new-class features. We also add a supervised contrastive term in the adapter space to promote intra-class compactness and old-new class separation. On Split CIFAR-100, our method improves over fine-tuning under Inc5, Inc10, and Inc20 without storing raw exemplars. The full Ours-LWM+Con model raises LastAcc from 4.55% to 31.64%, from 9.06% to 37.06%, and from 16.96% to 43.10% in Inc5, Inc10, and Inc20, respectively. It also achieves AvgAcc of 45.86%, 52.19%, and 56.18%. Ablation and retention analyses show that stable latent-state replay is the main source of the gain. Contrastive separation further refines the old-new geometry. These results suggest that prototype latent memory preserves reusable class-state distributions, rather than only fitting the current classifier.
Christian Zöllner, Mozzam Motiwala, Aysel Ahadova +4cs.CV cs.AI
Training of neural networks for histopathology classification tasks typically relies on data encoding into latent space, which reduces complexity and improves performance. There are several encoder networks available, either pretrained on general image datasets such as ImageNET, or specifically on histopathological images. Training of encoder networks should be adapted to downstream tasks, allowing encoding of biologic/diagnostic content while rendering networks invariant to label-irrelevant transformations. This paper investigates the effect of classical image transformation on the latent space, using networks provided by Lunit Inc. and Bioptimus, both focusing on pathological images, and by Meta Research Team. We assess variance of embeddings resulting from standard data transformations by comparing original and transformed image embeddings and by contrasting them with random, unrelated embeddings, using image tiles from hematoxylin/eosin-stained sections available in a colorectal tissue dataset and the publicly accessible TCGA dataset. Our findings show that embeddings of original and transformed images are closer to each other than to random embeddings, indicating robustness to transformations. However, they are not fully invariant, revealing that the encoder networks do not completely neutralize transformation effects in latent space, explaining why transformation-mediated augmentation of datasets can improve performance. Significant differences were observed between general and histopathology-specific encoder networks.
Despite the significant training acceleration and promising performance, Representation Autoencoders (RAEs) are mainly criticized for poor distillation effectiveness. In this work, we argue that RAE is competent at high-quality one-step generation. We achieve 1.48 FID with only 16-epoch distillation on ImageNet 256 dataset, surpassing various state-of-the-art methods. To achieve this, we quantitatively study the geometrical behavior of different underlying data spaces. We conclude that conventional distillation methods heavily rely on priors of plain teacher denoising trajectories, while RAE incurs much more complex trajectories with poor properties due to ill anisotropical latent space. We introduce the recently proposed drifting field as the distillation methodology, which makes use of semantically rich RAE latents and provides direct supervision involving no dependency. Bridging our Drift-RAE with previous generative paradigms, we propose several insightful modifications, including the first extrapolation-based guided sampling pipeline for one-step generation with barely no cost. The code will be made publicly available.
Existing 3D scene editing methods typically rely on per-scene optimization over explicit 3D representations or cascaded edit-and-reconstruct pipelines, resulting in high test-time cost, limited 3D awareness, and structural inconsistencies. To couple appearance synthesis and geometry prediction during editing, we build on a unified RGB-geometry reconstruction-generation latent space and adapt it to feed-forward 3D scene editing. The resulting framework, \textbf{JointEdit3D}, performs asymmetric latent inpainting by observing only a single edited RGB reference latent and generating the remaining RGB views and edited geometry latent under source-scene anchoring. JointEdit3D introduces a dedicated SceneAnchor Branch to inject source-scene structure without forcing direct copying, and adopts edit/background-aware losses to balance edited-region fidelity with unedited-content preservation. To address the lack of paired resources for standardized 3D scene editing evaluation, we introduce SceneEdit3D-15K, a dataset with 15K paired editing samples and renderer-provided 3D annotations, together with SceneEdit3D-Bench, a curated 100-sample benchmark. Experiments show that JointEdit3D improves edited-region quality and 3D structural completeness over prior baselines while maintaining competitive background preservation.
Video world models that maintain 3D spatial consistency across generated frames typically rely on explicit point cloud memory constructed in RGB space. This design is both computationally expensive, requiring repeated rendering and VAE encoding, and inherently lossy, as the round trip through pixel space discards rich features of the learned latent representation. In this paper, we introduce \emph{latent spatial memory} for video world models, a persistent 3D cache that stores scene information directly in the diffusion latent space, avoiding pixel-space reconstruction. Building on this, we propose Mirage, a latent-space spatial memory framework that constructs the memory by lifting latent tokens into 3D via depth-guided back-projection and queries it by synthesizing novel views through direct latent-space warping. This unified formulation eliminates both the information loss of pixel-space reconstruction and the computational burden of repeated encoding and rendering. Experiments show that latent spatial memory achieves up to \textbf{10.57}$\times$ faster end-to-end video generation and \textbf{55}$\times$ reduction in memory footprint relative to explicit 3D baselines. Leveraging the geometric prior of the diffusion model, Mirage attains state-of-the-art performance on WorldScore and strong reconstruction quality on RealEstate10K.
Won June Cho, Daeky Jeong, Hyeongyeol Lim +1cs.CV cs.AI cs.CE cs.LG
Synthetic histopathology image generation addresses critical challenges in computational pathology, including patient privacy and the growing need for large-scale training data for foundation models. Latent diffusion models have dominated the image generation domain, with recent works emphasizing that the choice of latent space is critical to the quality of generated images. Existing state-of-the-art generative models in histopathology use pretrained Vision Foundation Models (VFMs) as conditioning signals, and we observe that this leads to "conditioning collapse," where the conditioning signal dominates the latent space and lowers the quality and diversity of generated samples. Therefore, we instead use pretrained histopathology VFMs as the latent space itself, leveraging their patch-token features that encode rich semantic information. We empirically show that these features are $\ell_2$-normalized and lie on the unit hypersphere $\mathcal{S}^{d-1}$ with strong angular dominance and intrinsic curvature, making them naturally suited for a Riemannian formulation. We therefore present STREAM, the first framework to apply Riemannian flow matching in the pathology domain. STREAM consists of two stages: 1) a bridge-type stochastic perturbation that establishes per-token rectifiability on $\mathcal{S}^{d-1}$ for training a Diffusion Transformer (DiT) in latent space, and 2) a novel anisotropic decoder that allocates robustness to low-energy directions of the velocity-field Jacobian while preserving fidelity along its high-energy directions. Together, STREAM achieves state-of-the-art reconstruction and generation performance on breast and colorectal cancer datasets. The code will be publicly released upon acceptance.
Video Variational Autoencoder (VAE) enables latent video generative modeling by mapping the visual world into compact spatiotemporal latent spaces, improving training efficiency and stability. While existing video VAEs achieve commendable reconstruction quality, continued optimization of reconstruction does not necessarily translate into improved generative performance. How to enhance the diffusability of video latents remains a critical and unresolved challenge. In this work, inspired by principles of predictive world modeling, we investigate the potential of predictive learning to improve the video generative modeling. To this end, we introduce a simple and effective predictive reconstruction objective that unifies predictive learning with video reconstruction. Specifically, we randomly discard future frames and encode only partial past observations, while training the decoder to reconstruct the observed frames and predict future ones simultaneously. This design encourages the latent space to encode temporally predictive structures and build a more coherent understanding of video dynamics, thereby improving generation quality. Our model, termed Predictive Video VAE (PV-VAE), achieves superior performance on video generation, with 52% faster convergence and a 34.42 FVD improvement over the Wan2.2 VAE on UCF101. Furthermore, comprehensive analyses demonstrate that PV-VAE not only exhibits favorable scalability, with generative performance improving alongside VAE training, but also yields consistent gains in downstream video understanding, underscoring a latent space that effectively captures temporal coherence and motion priors.