#01Nov 19, 2026
cs.CV
Posterior Matching Distillation
Hidir Yesiltepe, Julius Berner, Arash Vahdat and 1 more
Diffusion distillation accelerates image and video generation by compressing a pretrained teacher's iterative sampling process into a few network evaluations. Distribution-based methods achieve this by matching the student's output distribution to the teacher's. This matching is often driven by reverse-KL objectives, whose mode-seeking bias favors dominant teacher modes at the expense of diversity across samples. We introduce Posterior Matching Distillation (PMD), a new member of the distribution-based distillation family that combines mode-covering posterior sampling with any-step generation. At zero population loss, PMD recovers the teacher's full posterior of clean images at each trained noise level. Its stochastic student supports any-step sampling budgets within the trained timestep grid without further training. To learn these posteriors, PMD matches exponentially reweighted averages of student samples conditioned on a fixed noisy state to corresponding teacher posterior means. Under Gaussian corruption, exponentially tilted posterior mean is a single teacher evaluation at a shifted input. We prove that, matching tilted means over an open neighborhood uniquely identifies the full posterior. PMD thus supervises the entire conditional distribution through a family of mean constraints, without any auxiliary student-score network. Experiments on SDXL and FLUX.2 [klein] 4B demonstrate that PMD achieves the best FID among 4-step methods on both models, while maintaining competitive or leading performance across prompt-alignment and perceptual-quality metrics.
#02Nov 19, 2026
cs.CV
Asuka-FM: A System-Architectural Co-Design Framework for Efficient Text-to-Image Pretraining on Consumer Hardware
Víctor Rodrigo González, Pablo M. Olmos, Ignacio Peis and 1 more
Frontier text-to-image (T2I) foundation models predominantly rely on multi-billion parameter Diffusion Transformers (DiTs) pretrained across massive compute clusters, creating an accessibility barrier for academic laboratories. When training DiTs from scratch on single consumer GPUs (e.g., an NVIDIA RTX 4090 with 24\,GB VRAM), practitioners encounter three primary bottlenecks: quadratic attention sequence costs, Just-In-Time (JIT) compilation graph breaks induced by variable aspect-ratio bucketing, and an autoencoder throughput wall where frozen Variational Autoencoder (VAE) operations dominate step latency. In this work, we introduce \textbf{Asuka-FM}, an end-to-end system and architectural framework demonstrating that competitive dual-stream DiTs ($\sim$206M parameters) can be pretrained from scratch on a single consumer GPU in $\sim$160 GPU hours for under \$55. Asuka-FM integrates four core designs: (i) an AdaLN-free dual-stream DiT backbone reducing parameter count; (ii) structured sequence sparsification via Sparse-Dense Residual Fusion (SPRINT) to cut attention FLOPs by over 50\%; (iii) continuous aspect-normalized Rotary Position Embeddings (RoPE) coupled with shifted square crop training, guaranteeing static tensor shapes while unlocking zero-shot virtual camera control; and (iv) a pixel adaptation phase utilizing a lightweight (10M parameter) Patch Detailer Head operating under velocity regression ($v$-prediction). Evaluated on open-domain benchmarks without benchmark-specific fine-tuning, Asuka-FM achieves GenEval 0.588 / DPG 78.96 scores at 256px and GenEval 0.609 / DPG 78.51 scores at 512px, setting a new Pareto-optimal efficiency standard for democratic foundation modeling.
#03Nov 19, 2026
cs.CV
One Prior, Two Dynamics: Repurposing Video Diffusion Transformers for Self-Supervised Latent Action Learning
Yuxin Jiang, Wenqiang Sun, Jiaqi Li and 4 more
Videos reveal how the world changes, but rarely the actions that cause those changes. Latent action learning recovers such supervision from visual transitions, yet often learns the underlying dynamics largely from scratch. We ask whether a pretrained video generator can instead provide the prior for both inverse and forward dynamics. We introduce O$^{2}$LaF, built on a simple principle: one prior, two dynamics. We initialize both models from the same pretrained video DiT and adapt them to their distinct roles: a feed-forward transition encoder with learnable action queries for inverse dynamics, and a causal, action-conditioned video generator for forward dynamics. By controlling information access and optimizing both dynamics end-to-end through a compact bottleneck, O$^{2}$LaF learns transferable actions from unlabeled videos. Across navigation and robotic manipulation, O$^{2}$LaF sets a new state of the art in few-shot action probing and enables stronger zero-shot action transfer across visual contexts and embodiments. Beyond generative video priors, we further extend our framework to predictive representation priors such as V-JEPA2, achieving competitive performance. Our findings suggest a broader paradigm for latent-action learning: repurposing pretrained video models to discover transferable actions and predict their consequences.
#04Nov 19, 2026
cs.CV
AdaPixFlow: Learning Continuous-Time 3D Dynamics Across Scenes from Sparse Multi-view Observations
Hrishikesh Viswanath, Aniket Bera
Understanding how objects move in dynamic scenes requires recovering their 3D motion from visual observations and predicting their positions at unobserved times. Existing methods either optimize scene-specific velocity fields or impose predefined physical and kinematic assumptions. We instead study learning a continuous-time dynamics operator across scenes, mapping multi-view visual observations to 3D object kinematics. We introduce Adaptive Pixel Flows (\ours{}), which learns this operator from multi-view RGB-D observations without per-scene optimization or predefined motion models. The method fuses observations from all cameras into a shared voxel grid in world coordinates, from which the network predicts a continuous-time velocity field. Because the operator is trained across scenes, it learns motion priors that carry over to unseen scenes. The field can then be integrated from observed states to predict object positions between observations and beyond the observed sequence. On held-out PhysInOne scenes, \ours{} improves F1 over the strongest competing method by 2.5\% for in-between prediction and 15\% for future prediction. Compared with per-scene velocity-field methods, the improvements are 30\% and 85\%, respectively. Applied zero-shot to Spring-Gaus, it improves future-prediction F1 by 12\%.
#05Nov 19, 2026
cs.CV
Deterministic Temporal Embedding for Calcium Imaging Trace Extraction
Qi Gao, Julius Beichert, Miriam Ratliff and 1 more
Reliable extraction of calcium-dependent fluorescence traces in calcium imaging sequences requires identifying pixels whose fluorescence intensity reflects the same underlying cellular activity. We present a deterministic temporal embedding framework that maps pixel-wise fluorescence traces to a low-dimensional representation in which temporally coherent sources form spatially coherent regions. Starting from a fuzzy-neighborhood graph and spectral initialization, we optimize a class-balanced cross-entropy objective using exact sparse attraction and a deterministic particle-mesh decomposition of the dense repulsive interaction with sparse edge compensation, avoiding stochastic negative sampling. The resulting latent representation is spatially decoded into reliable source supports for trace extraction without imposing an explicit cellular shape prior. Experiments on synthetic and in vivo two-photon calcium imaging data demonstrate robust recovery of irregular and spatially touching sources together with efficient deterministic embedding optimization.