Flow matching models have emerged as a powerful method for generative modeling on domains like images or videos, and even on irregular or unstructured data like 3D point clouds or even protein structures. These models are commonly trained in two stages: first, a data compressor is trained, and in a subsequent training stage a flow matching generative model is trained in the latent space of the data compressor. This two-stage paradigm sets obstacles for unifying models across data domains, as hand-crafted compressors architectures are used for different data modalities. To this end, we introduce INRFlow, a domain-agnostic approach to learn flow matching transformers directly in ambient space. Drawing inspiration from INRs, we introduce a conditionally independent point-wise training objective that enables INRFlow to make predictions continuously in coordinate space. Our empirical results demonstrate that INRFlow effectively handles different data modalities such as images, 3D point clouds and protein structure data, achieving strong performance in different domains and outperforming comparable approaches. INRFlow is a promising step towards domain-agnostic flow matching generative models that can be trivially adopted in different data domains.

Normalizing Flows (NFs) are likelihood-based models for continuous inputs. They have demonstrated promising results on both density estimation and generative modeling tasks, but have received relatively little attention in recent years. In this work, we demonstrate that NFs are more powerful than previously believed. We present TarFlow: a simple and scalable architecture that enables highly performant NF models. TarFlow can be thought of as a Transformer-based variant of Masked Autoregressive Flows (MAFs): it consists of a stack of autoregressive Transformer blocks on image patches, alternating the autoregression direction between layers. TarFlow is straightforward to train end-to-end, and capable of directly modeling and generating pixels. We also propose three key techniques to improve sample quality: Gaussian noise augmentation during training, a post training denoising procedure, and an effective guidance method for both class-conditional and unconditional settings. Putting these together, TarFlow sets new state-of-the-art results on likelihood estimation for images, beating the previous best methods by a large margin, and generates samples with quality and diversity comparable to diffusion models, for the first time with a stand-alone NF model. We make our code available at https://github.com/apple/ml-tarflow.

We present STARFlow, a scalable generative model based on normalizing flows that achieves strong performance on high-resolution image synthesis. STARFlow's main building block is Transformer Autoregressive Flow (TARFlow), which combines normalizing flows with Autoregressive Transformer architectures and has recently achieved impressive results in image modeling. In this work, we first establish the theoretical universality of TARFlow for modeling continuous distributions. Building on this foundation, we introduce a set of architectural and algorithmic innovations that significantly enhance the scalability: (1) a deep-shallow design where a deep Transformer block captures most of the model’s capacity, followed by a few shallow Transformer blocks that are computationally cheap yet contribute non-negligibly, (2) learning in the latent space of pretrained autoencoders, which proves far more effective than modeling pixels directly, and (3) a novel guidance algorithm that substantially improves sample quality. Crucially, our model remains a single, end-to-end normalizing flow, allowing exact maximum likelihood training in continuous space without discretization. STARFlow achieves competitive results in both class- and text-conditional image generation, with sample quality approaching that of state-of-the-art diffusion models. To our knowledge, this is the **first** successful demonstration of normalizing flows at this scale and resolution. Code and weights available at https://github.com/apple/ml-starflow.

Autoregressive models have driven remarkable progress in language modeling. Their foundational reliance on discrete tokens, unidirectional context, and single-pass decoding, while central to their success, also inspires the exploration of a design space that could offer new axes of modeling flexibility. In this work, we explore an alternative paradigm, shifting language modeling from a discrete token space to a continuous latent space. We propose a novel framework that employs transformer-based autoregressive normalizing flows to model these continuous representations. This approach unlocks substantial flexibility, enabling the construction of models that can capture global bi-directional context through stacked, alternating-direction autoregressive transformations, support block-wise generation with flexible token patch sizes, and facilitate a hierarchical multi-pass generation process. We further propose new mixture-based coupling transformations designed to capture complex dependencies within the latent space shaped by discrete data, and demonstrate theoretical connections to conventional discrete autoregressive models. Extensive experiments on language modeling benchmarks demonstrate strong likelihood performance and highlight the flexible modeling capabilities inherent in our framework.

Rendering scenes observed in a monocular video from novel viewpoints is a challenging problem. For static scenes the community has studied both scene-specific optimization techniques, which optimize on every test scene, and generalized techniques, which only run a deep net forward pass on a test scene. In contrast, for dynamic scenes, scene-specific optimization techniques exist, but, to our best knowledge, there is currently no generalized method for dynamic novel view synthesis from a given monocular video. To explore whether generalized dynamic novel view synthesis from monocular videos is possible today, we establish an analysis framework based on existing techniques and work toward the generalized approach. We find a pseudo-generalized process without scene-specific \emph{appearance} optimization is possible, but geometrically and temporally consistent depth estimates are needed. Despite no scene-specific appearance optimization, the pseudo-generalized approach improves upon some scene-specific methods.For more information see project page at https://xiaoming-zhao.github.io/projects/pgdvs.

We present Manifold Diffusion Fields (MDF), an approach that unlocks learning of diffusion models of data in general non-euclidean geometries. Leveraging insights from spectral geometry analysis, we define an intrinsic coordinate system on the manifold via the eigen-functions of the Laplace-Beltrami Operator. MDF represents functions using an explicit parametrization formed by a set of multiple input-output pairs. Our approach allows to sample continuous functions on manifolds and is invariant with respect to rigid and isometric transformations of the manifold. In addition, we show that MDF generalizes to the case where the training set contains functions on different manifolds. Empirical results on multiple datasets and manifolds including challenging scientific problems like weather prediction or molecular conformation show that MDF can capture distributions of such functions with better diversity and fidelity than previous approaches.

This paper introduces AIM, a collection of vision models pre-trained with an autoregressive objective. These models are inspired by their textual counterparts, i.e., Large Language Models (LLMs), and exhibit similar scaling properties. Specifically, we highlight two key findings: (1) the performance of the visual features scale with both the model capacity and the quantity of data, (2) the value of the objective function correlates with the performance of the model on downstream tasks. We illustrate the practical implication of these findings by pre-training a 7 billion parameter AIM on 2 billion images, that achieves 84.0% on ImageNet-1k with a frozen trunk. Interestingly, even at this scale, we observe no sign of saturation in performance, suggesting that AIM potentially represents a new frontier for training large-scale vision models. The pre-training of AIM is similar to the pre-training of LLMs, and does not require any image-specific strategy to stabilize the training at scale.

We present a novel way to predict molecular conformers through a simple formulation that sidesteps many of the heuristics of prior works and achieves state of the art results by using the advantages of scale. By training a diffusion generative model directly on 3D atomic positions without making assumptions about the explicit structure of molecules (e.g. modeling torsional angles) we are able to radically simplify structure learning, and make it trivial to scale up the model sizes. This model, called Molecular Conformer Fields (MCF), works by parameterizing conformer structures as functions that map elements from a molecular graph directly to their 3D location in space. This formulation allows us to boil down the essence of structure prediction to learning a distribution over functions. Experimental results show that scaling up the model capacity leads to large gains in generalization performance without enforcing inductive biases like rotational equivariance. MCF represents an advance in extending diffusion models to handle complex scientific problems in a conceptually simple, scalable and effective manner.

Reconstructing an accurate 3D object model from a few image observations remains a challenging problem in computer vision. State-of-the-art approaches typically assume accurate camera poses as input, which could be difficult to obtain in realistic settings. In this paper, we present FvOR, a learning-based object reconstruction method that predicts accurate 3D models given a few images with noisy input poses. The core of our approach is a fast and robust multi-view reconstruction algorithm to jointly refine 3D geometry and camera pose estimation using learnable neural network modules. We provide a thorough benchmark of state-of-the-art approaches for this problem on ShapeNet. Our approach achieves best-in-class results. It is also two orders of magnitude faster than the recent optimization-based approach IDR [67].

We introduce GAUDI, a generative model capable of capturing the distribution of complex and realistic 3D scenes that can be rendered immersively from a moving camera. We tackle this challenging problem with a scalable yet powerful approach, where we first optimize a latent representation that disentangles radiance fields and camera poses. This latent representation is then used to learn a generative model that enables both unconditional and conditional generation of 3D scenes. Our model generalizes previous works that focus on single objects by removing the assumption that the camera pose distribution can be shared across samples. We show that GAUDI obtains state-of-the-art performance in the unconditional generative setting across multiple datasets and allows for conditional generation of 3D scenes given conditioning variables like sparse image observations or text that describes the scene.

We study the problem of novel view synthesis from sparse source observations of a scene comprised of 3D objects. We propose a simple yet effective approach that is neither continuous nor implicit, challenging recent trends on view synthesis. Our approach explicitly encodes observations into a volumetric representation that enables amortized rendering. We demonstrate that although continuous radiance field representations have gained a lot of attention due to their expressive power, our simple approach obtains comparable or even better novel view reconstruction quality comparing with state-of-the-art baselines [49] while increasing rendering speed by over 400x. Our model is trained in a category-agnostic manner and does not require scene-specific optimization. Therefore, it is able to generalize novel view synthesis to object categories not seen during training. In addition, we show that with our simple formulation, we can use view synthesis as a self-supervision signal for efficient learning of 3D geometry without explicit 3D supervision.

We tackle the challenge of learning a distribution over complex, realistic, indoor scenes. In this paper, we introduce Generative Scene Networks (GSN), which learns to decompose scenes into a collection of many local radiance fields that can be rendered from a free moving camera. Our model can be used as a prior to generate new scenes, or to complete a scene given only sparse 2D observations. Recent work has shown that generative models of radiance fields can capture properties such as multi-view consistency and view-dependent lighting. However, these models are specialized for constrained viewing of single objects, such as cars or faces. Due to the size and complexity of realistic indoor environments, existing models lack the representational capacity to adequately capture them. Our decomposition scheme scales to larger and more complex scenes while preserving details and diversity, and the learned prior enables high-quality rendering from viewpoints that are significantly different from observed viewpoints. When compared to existing models, GSN produces quantitatively higher-quality scene renderings across several different scene datasets.

For many fundamental scene understanding tasks, it is difficult or impossible to obtain per-pixel ground truth labels from real images. We address this challenge by introducing Hypersim, a photorealistic synthetic dataset for holistic indoor scene understanding. To create our dataset, we leverage a large repository of synthetic scenes created by professional artists, and we generate 77,400 images of 461 indoor scenes with detailed per-pixel labels and corresponding ground truth geometry. Our dataset: (1) relies exclusively on publicly available 3D assets; (2) includes complete scene geometry, material information, and lighting information for every scene; (3) includes dense per-pixel semantic instance segmentations and complete camera information for every image; and (4) factors every image into diffuse reflectance, diffuse illumination, and a non-diffuse residual term that captures view-dependent lighting effects.We analyze our dataset at the level of scenes, objects, and pixels, and we analyze costs in terms of money, computation time, and annotation effort. Remarkably, we find that it is possible to generate our entire dataset from scratch, for roughly half the cost of training a popular open-source natural language processing model. We also evaluate sim-to-real transfer performance on two real-world scene understanding tasks – semantic segmentation and 3D shape prediction – where we find that pre-training on our dataset significantly improves performance on both tasks, and achieves state-of-the-art performance on the most challenging Pix3D test set. All of our rendered image data, as well as all the code we used to generate our dataset and perform our experiments, is available online.

State-of-the-art learning-based monocular 3D reconstruction methods learn priors over object categories on the training set, and as a result struggle to achieve reasonable generalization to object categories unseen during training. In this paper we study the inductive biases encoded in the model architecture that impact the generalization of learning-based 3D reconstruction methods. We find that 3 inductive biases impact performance: the spatial extent of the encoder, the use of the underlying geometry of the scene to describe point features, and the mechanism to aggregate information from multiple views. Additionally, we propose mechanisms to enforce those inductive biases: a point representation that is aware of camera position, and a variance cost to aggregate information across views. Our model achieves state-of-the-art results on the standard ShapeNet 3D reconstruction benchmark in various settings.

In most machine learning training paradigms a fixed, often handcrafted, loss function is assumed to be a good proxy for an underlying evaluation metric. In this work we assess this assumption by meta-learning an adaptive loss function to directly optimize the evaluation metric. We propose a sample efficient reinforcement learning approach for adapting the loss dynamically during training. We empirically show how this formulation improves performance by simultaneously optimizing the evaluation metric and smoothing the loss landscape. We verify our method in metric learning and classification scenarios, showing considerable improvements over the state-of-the-art on a diverse set of tasks. Importantly, our method is applicable to a wide range of loss functions and evaluation metrics. Furthermore, the learned policies are transferable across tasks and data, demonstrating the versatility of the method.

For an autonomous car, the ability to foresee a humans action is very useful for mitigating the risk of a possible collision. To humans this pedestrian intention foresight comes naturally as they are able to recognize another person's actions just by perceiving subtle changes in posture. Approximating this intention inference ability by directly training a deep neural network is useful but especially challenging. First, sufficiently large datasets for intention recognition with frame-wise human pose and intention annotations are rare and expensive to compile. Second, training on smaller datasets can lead to overfitting and make it difficult to adapt to intra-class variations in action executions. Therefore, in this paper, we propose a real time framework that learns (i) intention recognition using weak-supervision and (ii) locomotion dynamics of intention from pose information using transfer learning. This new formulation is able to tackle the lack of frame-wise annotations and to learn intra-class variation in action executions. We empirically demonstrate that our proposed approach leads to earlier and more stable detection of intention than other state of the art approaches with real time operation and the ability to detect intention one second before the pedestrian reaches the kerb.

Error Correcting Output Codes (ECOC) is a successful technique in multi-class classification, which is a core problem in Pattern Recognition and Machine Learning. A major advantage of ECOC over other methods is that the multi-class problem is decoupled into a set of binary problems that are solved independently. However, literature defines a general error-correcting capability for ECOCs without analyzing how it distributes among classes, hindering a deeper analysis of pair-wise error-correction. To address these limitations this paper proposes an Error-Correcting Factorization (ECF) method. Our contribution is three fold: (I) We propose a novel representation of the error-correction capability, called the design matrix, that enables us to build an ECOC on the basis of allocating correction to pairs of classes. (II) We derive the optimal code length of an ECOC using rank properties of the design matrix. (III) ECF is formulated as a discrete optimization problem, and a relaxed solution is found using an efficient constrained block coordinate descent approach. (IV) Enabled by the flexibility introduced with the design matrix we propose to allocate the error-correction on classes that are prone to confusion. Experimental results in several databases show that when allocating the error-correction to confusable classes ECF outperforms state-of-the-art approaches.

Unsupervised learning of visual similarities is of paramount importance to computer vision, particularly due to lacking training data for fine-grained similarities. Deep learning of similarities is often based on relationships between pairs or triplets of samples. Many of these relations are unreliable and mutually contradicting, implying inconsistencies when trained without supervision information that relates different tuples or triplets to each other. To overcome this problem, we use local estimates of reliable (dis-)similarities to initially group samples into compact surrogate classes and use local partial orders of samples to classes to link classes to each other. Similarity learning is then formulated as a partial ordering task with soft correspondences of all samples to classes. Adopting a strategy of self-supervision, a CNN is trained to optimally represent samples in a mutually consistent manner while updating the classes. The similarity learning and grouping procedure are integrated in a single model and optimized jointly. The proposed unsupervised approach shows competitive performance on detailed pose estimation and object classification.

Understanding human activity and being able to explain it in detail surpasses mere action classification by far in both complexity and value. The challenge is thus to describe an activity on the basis of its most fundamental constituents, the individual postures and their distinctive transitions. Supervised learning of such a fine-grained representation based on elementary poses is very tedious and does not scale. Therefore, we propose a completely unsupervised deep learning procedure based solely on video sequences, which starts from scratch without requiring pre-trained networks, predefined body models, or keypoints. A combinatorial sequence matching algorithm proposes relations between frames from subsets of the training data, while a CNN is reconciling the transitivity conflicts of the different subsets to learn a single concerted pose embedding despite changes in appearance across sequences. Without any manual annotation, the model learns a structured representation of postures and their temporal development. The model not only enables retrieval of similar postures but also temporal super-resolution. Additionally, based on a recurrent formulation, next frames can be synthesized.

Exemplar learning is a powerful paradigm for discovering visual similarities in an unsupervised manner. In this context, however, the recent breakthrough in deep learning could not yet unfold its full potential. With only a single positive sample, a great imbalance between one positive and many negatives, and unreliable relationships between most samples, training of convolutional neural networks is impaired. Given weak estimates of local distance we propose a single optimization problem to extract batches of samples with mutually consistent relations. Conflicting relations are distributed over different batches and similar samples are grouped into compact cliques. Learning exemplar similarities is framed as a sequence of clique categorization tasks. The CNN then consolidates transitivity relations within and between cliques and learns a single representation for all samples without the need for labels. The proposed unsupervised approach has shown competitive performance on detailed posture analysis and object classification.

We present an application of gesture recognition using an extension of dynamic time warping (DTW) to recognize behavioral patterns of attention deficit hyperactivity disorder (ADHD). We propose an extension of DTW using one-class classifiers in order to be able to encode the variability of a gesture category, and thus, perform an alignment between a gesture sample and a gesture class. We model the set of gesture samples of a certain gesture category using either Gaussian mixture models or an approximation of convex hulls. Thus, we add a theoretical contribution to classical warping path in DTW by including local modeling of intraclass gesture variability. This methodology is applied in a clinical context, detecting a group of ADHD behavioral patterns defined by experts in psychology/psychiatry, to provide support to clinicians in the diagnose procedure. The proposed methodology is tested on a novel multimodal dataset (RGB plus depth) of ADHD children recordings with behavioral patterns. We obtain satisfying results when compared to standard state-of-the-art approaches in the DTW context.