David J. Fleet

Klaus Greff, Francois Belletti, Lucas Beyer et al. · deepmind, mila

Data is the driving force of machine learning, with the amount and quality of training data often being more important for the performance of a system than architecture and training details. But collecting, processing and annotating real data at scale is difficult, expensive, and frequently raises additional privacy, fairness and legal concerns. Synthetic data is a powerful tool with the potential to address these shortcomings: 1) it is cheap 2) supports rich ground-truth annotations 3) offers full control over data and 4) can circumvent or mitigate problems regarding bias, privacy and licensing. Unfortunately, software tools for effective data generation are less mature than those for architecture design and training, which leads to fragmented generation efforts. To address these problems we introduce Kubric, an open-source Python framework that interfaces with PyBullet and Blender to generate photo-realistic scenes, with rich annotations, and seamlessly scales to large jobs distributed over thousands of machines, and generating TBs of data. We demonstrate the effectiveness of Kubric by presenting a series of 13 different generated datasets for tasks ranging from studying 3D NeRF models to optical flow estimation. We release Kubric, the used assets, all of the generation code, as well as the rendered datasets for reuse and modification.

Jonathan Ho, William Chan, Chitwan Saharia et al. · deepmind

We present Imagen Video, a text-conditional video generation system based on a cascade of video diffusion models. Given a text prompt, Imagen Video generates high definition videos using a base video generation model and a sequence of interleaved spatial and temporal video super-resolution models. We describe how we scale up the system as a high definition text-to-video model including design decisions such as the choice of fully-convolutional temporal and spatial super-resolution models at certain resolutions, and the choice of the v-parameterization of diffusion models. In addition, we confirm and transfer findings from previous work on diffusion-based image generation to the video generation setting. Finally, we apply progressive distillation to our video models with classifier-free guidance for fast, high quality sampling. We find Imagen Video not only capable of generating videos of high fidelity, but also having a high degree of controllability and world knowledge, including the ability to generate diverse videos and text animations in various artistic styles and with 3D object understanding. See https://imagen.research.google/video/ for samples.

Chitwan Saharia, William Chan, Saurabh Saxena et al. · deepmind

We present Imagen, a text-to-image diffusion model with an unprecedented degree of photorealism and a deep level of language understanding. Imagen builds on the power of large transformer language models in understanding text and hinges on the strength of diffusion models in high-fidelity image generation. Our key discovery is that generic large language models (e.g. T5), pretrained on text-only corpora, are surprisingly effective at encoding text for image synthesis: increasing the size of the language model in Imagen boosts both sample fidelity and image-text alignment much more than increasing the size of the image diffusion model. Imagen achieves a new state-of-the-art FID score of 7.27 on the COCO dataset, without ever training on COCO, and human raters find Imagen samples to be on par with the COCO data itself in image-text alignment. To assess text-to-image models in greater depth, we introduce DrawBench, a comprehensive and challenging benchmark for text-to-image models. With DrawBench, we compare Imagen with recent methods including VQ-GAN+CLIP, Latent Diffusion Models, and DALL-E 2, and find that human raters prefer Imagen over other models in side-by-side comparisons, both in terms of sample quality and image-text alignment. See https://imagen.research.google/ for an overview of the results.

Sara Sabour, Suhani Vora, Daniel Duckworth et al. · deepmind

Neural radiance fields (NeRF) excel at synthesizing new views given multi-view, calibrated images of a static scene. When scenes include distractors, which are not persistent during image capture (moving objects, lighting variations, shadows), artifacts appear as view-dependent effects or 'floaters'. To cope with distractors, we advocate a form of robust estimation for NeRF training, modeling distractors in training data as outliers of an optimization problem. Our method successfully removes outliers from a scene and improves upon our baselines, on synthetic and real-world scenes. Our technique is simple to incorporate in modern NeRF frameworks, with few hyper-parameters. It does not assume a priori knowledge of the types of distractors, and is instead focused on the optimization problem rather than pre-processing or modeling transient objects. More results on our page https://robustnerf.github.io.

Renjie Liao, Simon Kornblith, Mengye Ren et al.

We revisit the challenging problem of training Gaussian-Bernoulli restricted Boltzmann machines (GRBMs), introducing two innovations. We propose a novel Gibbs-Langevin sampling algorithm that outperforms existing methods like Gibbs sampling. We propose a modified contrastive divergence (CD) algorithm so that one can generate images with GRBMs starting from noise. This enables direct comparison of GRBMs with deep generative models, improving evaluation protocols in the RBM literature. Moreover, we show that modified CD and gradient clipping are enough to robustly train GRBMs with large learning rates, thus removing the necessity of various tricks in the literature. Experiments on Gaussian Mixtures, MNIST, FashionMNIST, and CelebA show GRBMs can generate good samples, despite their single-hidden-layer architecture. Our code is released at: \url{https://github.com/lrjconan/GRBM}.

Shayan Shekarforoush, David B. Lindell, David J. Fleet et al. · stanford

Coordinate networks like Multiplicative Filter Networks (MFNs) and BACON offer some control over the frequency spectrum used to represent continuous signals such as images or 3D volumes. Yet, they are not readily applicable to problems for which coarse-to-fine estimation is required, including various inverse problems in which coarse-to-fine optimization plays a key role in avoiding poor local minima. We introduce a new coordinate network architecture and training scheme that enables coarse-to-fine optimization with fine-grained control over the frequency support of learned reconstructions. This is achieved with two key innovations. First, we incorporate skip connections so that structure at one scale is preserved when fitting finer-scale structure. Second, we propose a novel initialization scheme to provide control over the model frequency spectrum at each stage of optimization. We demonstrate how these modifications enable multiscale optimization for coarse-to-fine fitting to natural images. We then evaluate our model on synthetically generated datasets for the the problem of single-particle cryo-EM reconstruction. We learn high resolution multiscale structures, on par with the state-of-the art.

Jonathan Ho, Tim Salimans, Alexey Gritsenko et al.

Generating temporally coherent high fidelity video is an important milestone in generative modeling research. We make progress towards this milestone by proposing a diffusion model for video generation that shows very promising initial results. Our model is a natural extension of the standard image diffusion architecture, and it enables jointly training from image and video data, which we find to reduce the variance of minibatch gradients and speed up optimization. To generate long and higher resolution videos we introduce a new conditional sampling technique for spatial and temporal video extension that performs better than previously proposed methods. We present the first results on a large text-conditioned video generation task, as well as state-of-the-art results on established benchmarks for video prediction and unconditional video generation. Supplementary material is available at https://video-diffusion.github.io/

Tao Tu, Shekoofeh Azizi, Danny Driess et al.

Medicine is inherently multimodal, with rich data modalities spanning text, imaging, genomics, and more. Generalist biomedical artificial intelligence (AI) systems that flexibly encode, integrate, and interpret this data at scale can potentially enable impactful applications ranging from scientific discovery to care delivery. To enable the development of these models, we first curate MultiMedBench, a new multimodal biomedical benchmark. MultiMedBench encompasses 14 diverse tasks such as medical question answering, mammography and dermatology image interpretation, radiology report generation and summarization, and genomic variant calling. We then introduce Med-PaLM Multimodal (Med-PaLM M), our proof of concept for a generalist biomedical AI system. Med-PaLM M is a large multimodal generative model that flexibly encodes and interprets biomedical data including clinical language, imaging, and genomics with the same set of model weights. Med-PaLM M reaches performance competitive with or exceeding the state of the art on all MultiMedBench tasks, often surpassing specialist models by a wide margin. We also report examples of zero-shot generalization to novel medical concepts and tasks, positive transfer learning across tasks, and emergent zero-shot medical reasoning. To further probe the capabilities and limitations of Med-PaLM M, we conduct a radiologist evaluation of model-generated (and human) chest X-ray reports and observe encouraging performance across model scales. In a side-by-side ranking on 246 retrospective chest X-rays, clinicians express a pairwise preference for Med-PaLM M reports over those produced by radiologists in up to 40.50% of cases, suggesting potential clinical utility. While considerable work is needed to validate these models in real-world use cases, our results represent a milestone towards the development of generalist biomedical AI systems.

Shekoofeh Azizi, Simon Kornblith, Chitwan Saharia et al.

Deep generative models are becoming increasingly powerful, now generating diverse high fidelity photo-realistic samples given text prompts. Have they reached the point where models of natural images can be used for generative data augmentation, helping to improve challenging discriminative tasks? We show that large-scale text-to image diffusion models can be fine-tuned to produce class conditional models with SOTA FID (1.76 at 256x256 resolution) and Inception Score (239 at 256x256). The model also yields a new SOTA in Classification Accuracy Scores (64.96 for 256x256 generative samples, improving to 69.24 for 1024x1024 samples). Augmenting the ImageNet training set with samples from the resulting models yields significant improvements in ImageNet classification accuracy over strong ResNet and Vision Transformer baselines.

Kevin Clark, Paul Vicol, Kevin Swersky et al.

We present Direct Reward Fine-Tuning (DRaFT), a simple and effective method for fine-tuning diffusion models to maximize differentiable reward functions, such as scores from human preference models. We first show that it is possible to backpropagate the reward function gradient through the full sampling procedure, and that doing so achieves strong performance on a variety of rewards, outperforming reinforcement learning-based approaches. We then propose more efficient variants of DRaFT: DRaFT-K, which truncates backpropagation to only the last K steps of sampling, and DRaFT-LV, which obtains lower-variance gradient estimates for the case when K=1. We show that our methods work well for a variety of reward functions and can be used to substantially improve the aesthetic quality of images generated by Stable Diffusion 1.4. Finally, we draw connections between our approach and prior work, providing a unifying perspective on the design space of gradient-based fine-tuning algorithms.

Ting Chen, Saurabh Saxena, Lala Li et al.

While language tasks are naturally expressed in a single, unified, modeling framework, i.e., generating sequences of tokens, this has not been the case in computer vision. As a result, there is a proliferation of distinct architectures and loss functions for different vision tasks. In this work we show that a diverse set of "core" computer vision tasks can also be unified if formulated in terms of a shared pixel-to-sequence interface. We focus on four tasks, namely, object detection, instance segmentation, keypoint detection, and image captioning, all with diverse types of outputs, e.g., bounding boxes or dense masks. Despite that, by formulating the output of each task as a sequence of discrete tokens with a unified interface, we show that one can train a neural network with a single model architecture and loss function on all these tasks, with no task-specific customization. To solve a specific task, we use a short prompt as task description, and the sequence output adapts to the prompt so it can produce task-specific output. We show that such a model can achieve competitive performance compared to well-established task-specific models.

Su Wang, Chitwan Saharia, Ceslee Montgomery et al.

Text-guided image editing can have a transformative impact in supporting creative applications. A key challenge is to generate edits that are faithful to input text prompts, while consistent with input images. We present Imagen Editor, a cascaded diffusion model built, by fine-tuning Imagen on text-guided image inpainting. Imagen Editor's edits are faithful to the text prompts, which is accomplished by using object detectors to propose inpainting masks during training. In addition, Imagen Editor captures fine details in the input image by conditioning the cascaded pipeline on the original high resolution image. To improve qualitative and quantitative evaluation, we introduce EditBench, a systematic benchmark for text-guided image inpainting. EditBench evaluates inpainting edits on natural and generated images exploring objects, attributes, and scenes. Through extensive human evaluation on EditBench, we find that object-masking during training leads to across-the-board improvements in text-image alignment -- such that Imagen Editor is preferred over DALL-E 2 and Stable Diffusion -- and, as a cohort, these models are better at object-rendering than text-rendering, and handle material/color/size attributes better than count/shape attributes.

Ting Chen, Lala Li, Saurabh Saxena et al.

Panoptic segmentation assigns semantic and instance ID labels to every pixel of an image. As permutations of instance IDs are also valid solutions, the task requires learning of high-dimensional one-to-many mapping. As a result, state-of-the-art approaches use customized architectures and task-specific loss functions. We formulate panoptic segmentation as a discrete data generation problem, without relying on inductive bias of the task. A diffusion model is proposed to model panoptic masks, with a simple architecture and generic loss function. By simply adding past predictions as a conditioning signal, our method is capable of modeling video (in a streaming setting) and thereby learns to track object instances automatically. With extensive experiments, we demonstrate that our simple approach can perform competitively to state-of-the-art specialist methods in similar settings.

Allan Jabri, David Fleet, Ting Chen

Natural data is redundant yet predominant architectures tile computation uniformly across their input and output space. We propose the Recurrent Interface Networks (RINs), an attention-based architecture that decouples its core computation from the dimensionality of the data, enabling adaptive computation for more scalable generation of high-dimensional data. RINs focus the bulk of computation (i.e. global self-attention) on a set of latent tokens, using cross-attention to read and write (i.e. route) information between latent and data tokens. Stacking RIN blocks allows bottom-up (data to latent) and top-down (latent to data) feedback, leading to deeper and more expressive routing. While this routing introduces challenges, this is less problematic in recurrent computation settings where the task (and routing problem) changes gradually, such as iterative generation with diffusion models. We show how to leverage recurrence by conditioning the latent tokens at each forward pass of the reverse diffusion process with those from prior computation, i.e. latent self-conditioning. RINs yield state-of-the-art pixel diffusion models for image and video generation, scaling to 1024X1024 images without cascades or guidance, while being domain-agnostic and up to 10X more efficient than 2D and 3D U-Nets.

Shekoofeh Azizi, Laura Culp, Jan Freyberg et al.

Recent progress in Medical Artificial Intelligence (AI) has delivered systems that can reach clinical expert level performance. However, such systems tend to demonstrate sub-optimal "out-of-distribution" performance when evaluated in clinical settings different from the training environment. A common mitigation strategy is to develop separate systems for each clinical setting using site-specific data [1]. However, this quickly becomes impractical as medical data is time-consuming to acquire and expensive to annotate [2]. Thus, the problem of "data-efficient generalization" presents an ongoing difficulty for Medical AI development. Although progress in representation learning shows promise, their benefits have not been rigorously studied, specifically for out-of-distribution settings. To meet these challenges, we present REMEDIS, a unified representation learning strategy to improve robustness and data-efficiency of medical imaging AI. REMEDIS uses a generic combination of large-scale supervised transfer learning with self-supervised learning and requires little task-specific customization. We study a diverse range of medical imaging tasks and simulate three realistic application scenarios using retrospective data. REMEDIS exhibits significantly improved in-distribution performance with up to 11.5% relative improvement in diagnostic accuracy over a strong supervised baseline. More importantly, our strategy leads to strong data-efficient generalization of medical imaging AI, matching strong supervised baselines using between 1% to 33% of retraining data across tasks. These results suggest that REMEDIS can significantly accelerate the life-cycle of medical imaging AI development thereby presenting an important step forward for medical imaging AI to deliver broad impact.

Saurabh Saxena, Charles Herrmann, Junhwa Hur et al.

Denoising diffusion probabilistic models have transformed image generation with their impressive fidelity and diversity. We show that they also excel in estimating optical flow and monocular depth, surprisingly, without task-specific architectures and loss functions that are predominant for these tasks. Compared to the point estimates of conventional regression-based methods, diffusion models also enable Monte Carlo inference, e.g., capturing uncertainty and ambiguity in flow and depth. With self-supervised pre-training, the combined use of synthetic and real data for supervised training, and technical innovations (infilling and step-unrolled denoising diffusion training) to handle noisy-incomplete training data, and a simple form of coarse-to-fine refinement, one can train state-of-the-art diffusion models for depth and optical flow estimation. Extensive experiments focus on quantitative performance against benchmarks, ablations, and the model's ability to capture uncertainty and multimodality, and impute missing values. Our model, DDVM (Denoising Diffusion Vision Model), obtains a state-of-the-art relative depth error of 0.074 on the indoor NYU benchmark and an Fl-all outlier rate of 3.26\% on the KITTI optical flow benchmark, about 25\% better than the best published method. For an overview see https://diffusion-vision.github.io.

Saurabh Saxena, Abhishek Kar, Mohammad Norouzi et al.

We formulate monocular depth estimation using denoising diffusion models, inspired by their recent successes in high fidelity image generation. To that end, we introduce innovations to address problems arising due to noisy, incomplete depth maps in training data, including step-unrolled denoising diffusion, an $L_1$ loss, and depth infilling during training. To cope with the limited availability of data for supervised training, we leverage pre-training on self-supervised image-to-image translation tasks. Despite the simplicity of the approach, with a generic loss and architecture, our DepthGen model achieves SOTA performance on the indoor NYU dataset, and near SOTA results on the outdoor KITTI dataset. Further, with a multimodal posterior, DepthGen naturally represents depth ambiguity (e.g., from transparent surfaces), and its zero-shot performance combined with depth imputation, enable a simple but effective text-to-3D pipeline. Project page: https://depth-gen.github.io

Deqing Sun, Charles Herrmann, Fitsum Reda et al.

How important are training details and datasets to recent optical flow models like RAFT? And do they generalize? To explore these questions, rather than develop a new model, we revisit three prominent models, PWC-Net, IRR-PWC and RAFT, with a common set of modern training techniques and datasets, and observe significant performance gains, demonstrating the importance and generality of these training details. Our newly trained PWC-Net and IRR-PWC models show surprisingly large improvements, up to 30% versus original published results on Sintel and KITTI 2015 benchmarks. They outperform the more recent Flow1D on KITTI 2015 while being 3x faster during inference. Our newly trained RAFT achieves an Fl-all score of 4.31% on KITTI 2015, more accurate than all published optical flow methods at the time of writing. Our results demonstrate the benefits of separating the contributions of models, training techniques and datasets when analyzing performance gains of optical flow methods. Our source code will be publicly available.

Hshmat Sahak, Daniel Watson, Chitwan Saharia et al.

Diffusion models have shown promising results on single-image super-resolution and other image- to-image translation tasks. Despite this success, they have not outperformed state-of-the-art GAN models on the more challenging blind super-resolution task, where the input images are out of distribution, with unknown degradations. This paper introduces SR3+, a diffusion-based model for blind super-resolution, establishing a new state-of-the-art. To this end, we advocate self-supervised training with a combination of composite, parameterized degradations for self-supervised training, and noise-conditioing augmentation during training and testing. With these innovations, a large-scale convolutional architecture, and large-scale datasets, SR3+ greatly outperforms SR3. It outperforms Real-ESRGAN when trained on the same data, with a DRealSR FID score of 36.82 vs. 37.22, which further improves to FID of 32.37 with larger models, and further still with larger training sets.

Daniel Watson, Saurabh Saxena, Lala Li et al.

We present 4DiM, a cascaded diffusion model for 4D novel view synthesis (NVS), supporting generation with arbitrary camera trajectories and timestamps, in natural scenes, conditioned on one or more images. With a novel architecture and sampling procedure, we enable training on a mixture of 3D (with camera pose), 4D (pose+time) and video (time but no pose) data, which greatly improves generalization to unseen images and camera pose trajectories over prior works that focus on limited domains (e.g., object centric). 4DiM is the first-ever NVS method with intuitive metric-scale camera pose control enabled by our novel calibration pipeline for structure-from-motion-posed data. Experiments demonstrate that 4DiM outperforms prior 3D NVS models both in terms of image fidelity and pose alignment, while also enabling the generation of scene dynamics. 4DiM provides a general framework for a variety of tasks including single-image-to-3D, two-image-to-video (interpolation and extrapolation), and pose-conditioned video-to-video translation, which we illustrate qualitatively on a variety of scenes. For an overview see https://4d-diffusion.github.io

Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri et al.

Artificial intelligence (AI) has significant potential in healthcare applications, but its training and deployment faces challenges due to healthcare's diverse data, complex tasks, and the need to preserve privacy. Foundation models that perform well on medical tasks and require less task-specific tuning data are critical to accelerate the development of healthcare AI applications. We introduce MedGemma, a collection of medical vision-language foundation models based on Gemma 3 4B and 27B. MedGemma demonstrates advanced medical understanding and reasoning on images and text, significantly exceeding the performance of similar-sized generative models and approaching the performance of task-specific models, while maintaining the general capabilities of the Gemma 3 base models. For out-of-distribution tasks, MedGemma achieves 2.6-10% improvement on medical multimodal question answering, 15.5-18.1% improvement on chest X-ray finding classification, and 10.8% improvement on agentic evaluations compared to the base models. Fine-tuning MedGemma further improves performance in subdomains, reducing errors in electronic health record information retrieval by 50% and reaching comparable performance to existing specialized state-of-the-art methods for pneumothorax classification and histopathology patch classification. We additionally introduce MedSigLIP, a medically-tuned vision encoder derived from SigLIP. MedSigLIP powers the visual understanding capabilities of MedGemma and as an encoder achieves comparable or better performance than specialized medical image encoders. Taken together, the MedGemma collection provides a strong foundation of medical image and text capabilities, with potential to significantly accelerate medical research and development of downstream applications. The MedGemma collection, including tutorials and model weights, can be found at https://goo.gle/medgemma.

Saurabh Saxena, Junhwa Hur, Charles Herrmann et al.

While methods for monocular depth estimation have made significant strides on standard benchmarks, zero-shot metric depth estimation remains unsolved. Challenges include the joint modeling of indoor and outdoor scenes, which often exhibit significantly different distributions of RGB and depth, and the depth-scale ambiguity due to unknown camera intrinsics. Recent work has proposed specialized multi-head architectures for jointly modeling indoor and outdoor scenes. In contrast, we advocate a generic, task-agnostic diffusion model, with several advancements such as log-scale depth parameterization to enable joint modeling of indoor and outdoor scenes, conditioning on the field-of-view (FOV) to handle scale ambiguity and synthetically augmenting FOV during training to generalize beyond the limited camera intrinsics in training datasets. Furthermore, by employing a more diverse training mixture than is common, and an efficient diffusion parameterization, our method, DMD (Diffusion for Metric Depth) achieves a 25\% reduction in relative error (REL) on zero-shot indoor and 33\% reduction on zero-shot outdoor datasets over the current SOTA using only a small number of denoising steps. For an overview see https://diffusion-vision.github.io/dmd

Eric Wang, Samuel Schmidgall, Paul F. Jaeger et al.

Therapeutic development is a costly and high-risk endeavor that is often plagued by high failure rates. To address this, we introduce TxGemma, a suite of efficient, generalist large language models (LLMs) capable of therapeutic property prediction as well as interactive reasoning and explainability. Unlike task-specific models, TxGemma synthesizes information from diverse sources, enabling broad application across the therapeutic development pipeline. The suite includes 2B, 9B, and 27B parameter models, fine-tuned from Gemma-2 on a comprehensive dataset of small molecules, proteins, nucleic acids, diseases, and cell lines. Across 66 therapeutic development tasks, TxGemma achieved superior or comparable performance to the state-of-the-art generalist model on 64 (superior on 45), and against state-of-the-art specialist models on 50 (superior on 26). Fine-tuning TxGemma models on therapeutic downstream tasks, such as clinical trial adverse event prediction, requires less training data than fine-tuning base LLMs, making TxGemma suitable for data-limited applications. Beyond these predictive capabilities, TxGemma features conversational models that bridge the gap between general LLMs and specialized property predictors. These allow scientists to interact in natural language, provide mechanistic reasoning for predictions based on molecular structure, and engage in scientific discussions. Building on this, we further introduce Agentic-Tx, a generalist therapeutic agentic system powered by Gemini 2.5 that reasons, acts, manages diverse workflows, and acquires external domain knowledge. Agentic-Tx surpasses prior leading models on the Humanity's Last Exam benchmark (Chemistry & Biology) with 52.3% relative improvement over o3-mini (high) and 26.7% over o3-mini (high) on GPQA (Chemistry) and excels with improvements of 6.3% (ChemBench-Preference) and 2.4% (ChemBench-Mini) over o3-mini (high).

Lily Goli, Sara Sabour, Mark Matthews et al.

There has been extensive progress in the reconstruction and generation of 4D scenes from monocular casually-captured video. While these tasks rely heavily on known camera poses, the problem of finding such poses using structure-from-motion (SfM) often depends on robustly separating static from dynamic parts of a video. The lack of a robust solution to this problem limits the performance of SfM camera-calibration pipelines. We propose a novel approach to video-based motion segmentation to identify the components of a scene that are moving w.r.t. a fixed world frame. Our simple but effective iterative method, RoMo, combines optical flow and epipolar cues with a pre-trained video segmentation model. It outperforms unsupervised baselines for motion segmentation as well as supervised baselines trained from synthetic data. More importantly, the combination of an off-the-shelf SfM pipeline with our segmentation masks establishes a new state-of-the-art on camera calibration for scenes with dynamic content, outperforming existing methods by a substantial margin.

Junhwa Hur, Charles Herrmann, Saurabh Saxena et al.

Despite the recent progress, existing frame interpolation methods still struggle with processing extremely high resolution input and handling challenging cases such as repetitive textures, thin objects, and large motion. To address these issues, we introduce a patch-based cascaded pixel diffusion model for high resolution frame interpolation, HiFI, that excels in these scenarios while achieving competitive performance on standard benchmarks. Cascades, which generate a series of images from low to high resolution, can help significantly with large or complex motion that require both global context for a coarse solution and detailed context for high resolution output. However, contrary to prior work on cascaded diffusion models which perform diffusion on increasingly large resolutions, we use a single model that always performs diffusion at the same resolution and upsamples by processing patches of the inputs and the prior solution. At inference time, this drastically reduces memory usage and allows a single model, solving both frame interpolation (base model's task) and spatial up-sampling, saving training cost as well. HiFI excels at high-resolution images and complex repeated textures that require global context, achieving comparable or state-of-the-art performance on various benchmarks (Vimeo, Xiph, X-Test, and SEPE-8K). We further introduce a new dataset, LaMoR, that focuses on particularly challenging cases, and HiFI significantly outperforms other baselines. Please visit our project page for video results: https://hifi-diffusion.github.io

Farzaneh Mahdisoltani, Saeed Mahdisoltani, Roger B. Grosse et al.

Understanding the internal representations and decision mechanisms of deep neural networks remains a critical open challenge. While existing interpretability methods often identify influential input regions, they may not elucidate how a model distinguishes between classes or what specific changes would transition an input from one category to another. To address these limitations, we propose a novel framework that visualizes the implicit path between classes by treating the network gradient as a form of infinitesimal motion. Drawing inspiration from phase-based motion magnification, we first decompose images using invertible transforms-specifically the Complex Steerable Pyramid-then compute class-conditional gradients in the transformed space. Rather than iteratively integrating the gradient to trace a full path, we amplify the one-step gradient to the input and perform a linear extrapolation to expose how the model moves from source to target class. By operating in the steerable pyramid domain, these amplified gradients produce semantically meaningful, spatially coherent morphs that highlight the classifier's most sensitive directions, giving insight into the geometry of its decision boundaries. Experiments on both synthetic and real-world datasets demonstrate that our phase-focused extrapolation yields perceptually aligned, semantically meaningful transformations, offering a novel, interpretable lens into neural classifiers' internal representations.

Sara Sabour, Lily Goli, George Kopanas et al.

3D Gaussian Splatting (3DGS) is a promising technique for 3D reconstruction, offering efficient training and rendering speeds, making it suitable for real-time applications.However, current methods require highly controlled environments (no moving people or wind-blown elements, and consistent lighting) to meet the inter-view consistency assumption of 3DGS. This makes reconstruction of real-world captures problematic. We present SpotLessSplats, an approach that leverages pre-trained and general-purpose features coupled with robust optimization to effectively ignore transient distractors. Our method achieves state-of-the-art reconstruction quality both visually and quantitatively, on casual captures. Additional results available at: https://spotlesssplats.github.io

Shayan Shekarforoush, David B. Lindell, Marcus A. Brubaker et al.

Cryo-EM is an increasingly popular method for determining the atomic resolution 3D structure of macromolecular complexes (eg, proteins) from noisy 2D images captured by an electron microscope. The computational task is to reconstruct the 3D density of the particle, along with 3D pose of the particle in each 2D image, for which the posterior pose distribution is highly multi-modal. Recent developments in cryo-EM have focused on deep learning for which amortized inference has been used to predict pose. Here, we address key problems with this approach, and propose a new semi-amortized method, cryoSPIN, in which reconstruction begins with amortized inference and then switches to a form of auto-decoding to refine poses locally using stochastic gradient descent. Through evaluation on synthetic datasets, we demonstrate that cryoSPIN is able to handle multi-modal pose distributions during the amortized inference stage, while the later, more flexible stage of direct pose optimization yields faster and more accurate convergence of poses compared to baselines. On experimental data, we show that cryoSPIN outperforms the state-of-the-art cryoAI in speed and reconstruction quality.

Juan Manuel Zambrano Chaves, Eric Wang, Tao Tu et al.

Developing therapeutics is a lengthy and expensive process that requires the satisfaction of many different criteria, and AI models capable of expediting the process would be invaluable. However, the majority of current AI approaches address only a narrowly defined set of tasks, often circumscribed within a particular domain. To bridge this gap, we introduce Tx-LLM, a generalist large language model (LLM) fine-tuned from PaLM-2 which encodes knowledge about diverse therapeutic modalities. Tx-LLM is trained using a collection of 709 datasets that target 66 tasks spanning various stages of the drug discovery pipeline. Using a single set of weights, Tx-LLM simultaneously processes a wide variety of chemical or biological entities(small molecules, proteins, nucleic acids, cell lines, diseases) interleaved with free-text, allowing it to predict a broad range of associated properties, achieving competitive with state-of-the-art (SOTA) performance on 43 out of 66 tasks and exceeding SOTA on 22. Among these, Tx-LLM is particularly powerful and exceeds best-in-class performance on average for tasks combining molecular SMILES representations with text such as cell line names or disease names, likely due to context learned during pretraining. We observe evidence of positive transfer between tasks with diverse drug types (e.g.,tasks involving small molecules and tasks involving proteins), and we study the impact of model size, domain finetuning, and prompting strategies on performance. We believe Tx-LLM represents an important step towards LLMs encoding biochemical knowledge and could have a future role as an end-to-end tool across the drug discovery development pipeline.

Lin Yang, Shawn Xu, Andrew Sellergren et al.

Many clinical tasks require an understanding of specialized data, such as medical images and genomics, which is not typically found in general-purpose large multimodal models. Building upon Gemini's multimodal models, we develop several models within the new Med-Gemini family that inherit core capabilities of Gemini and are optimized for medical use via fine-tuning with 2D and 3D radiology, histopathology, ophthalmology, dermatology and genomic data. Med-Gemini-2D sets a new standard for AI-based chest X-ray (CXR) report generation based on expert evaluation, exceeding previous best results across two separate datasets by an absolute margin of 1% and 12%, where 57% and 96% of AI reports on normal cases, and 43% and 65% on abnormal cases, are evaluated as "equivalent or better" than the original radiologists' reports. We demonstrate the first ever large multimodal model-based report generation for 3D computed tomography (CT) volumes using Med-Gemini-3D, with 53% of AI reports considered clinically acceptable, although additional research is needed to meet expert radiologist reporting quality. Beyond report generation, Med-Gemini-2D surpasses the previous best performance in CXR visual question answering (VQA) and performs well in CXR classification and radiology VQA, exceeding SoTA or baselines on 17 of 20 tasks. In histopathology, ophthalmology, and dermatology image classification, Med-Gemini-2D surpasses baselines across 18 out of 20 tasks and approaches task-specific model performance. Beyond imaging, Med-Gemini-Polygenic outperforms the standard linear polygenic risk score-based approach for disease risk prediction and generalizes to genetically correlated diseases for which it has never been trained. Although further development and evaluation are necessary in the safety-critical medical domain, our results highlight the potential of Med-Gemini across a wide range of medical tasks.

Mehdy Dousty, David J. Fleet, José Zariffa

Hand function is critical for our interactions and quality of life. Spinal cord injuries (SCI) can impair hand function, reducing independence. A comprehensive evaluation of function in home and community settings requires a hand grasp taxonomy for individuals with impaired hand function. Developing such a taxonomy is challenging due to unrepresented grasp types in standard taxonomies, uneven data distribution across injury levels, and limited data. This study aims to automatically identify the dominant distinct hand grasps in egocentric video using semantic clustering. Egocentric video recordings collected in the homes of 19 individual with cervical SCI were used to cluster grasping actions with semantic significance. A deep learning model integrating posture and appearance data was employed to create a personalized hand taxonomy. Quantitative analysis reveals a cluster purity of 67.6% +- 24.2% with with 18.0% +- 21.8% redundancy. Qualitative assessment revealed meaningful clusters in video content. This methodology provides a flexible and effective strategy to analyze hand function in the wild. It offers researchers and clinicians an efficient tool for evaluating hand function, aiding sensitive assessments and tailored intervention plans.

Ting Chen, Saurabh Saxena, Lala Li et al.

We present Pix2Seq, a simple and generic framework for object detection. Unlike existing approaches that explicitly integrate prior knowledge about the task, we cast object detection as a language modeling task conditioned on the observed pixel inputs. Object descriptions (e.g., bounding boxes and class labels) are expressed as sequences of discrete tokens, and we train a neural network to perceive the image and generate the desired sequence. Our approach is based mainly on the intuition that if a neural network knows about where and what the objects are, we just need to teach it how to read them out. Beyond the use of task-specific data augmentations, our approach makes minimal assumptions about the task, yet it achieves competitive results on the challenging COCO dataset, compared to highly specialized and well optimized detection algorithms.

Jonathan Ho, Chitwan Saharia, William Chan et al.

We show that cascaded diffusion models are capable of generating high fidelity images on the class-conditional ImageNet generation benchmark, without any assistance from auxiliary image classifiers to boost sample quality. A cascaded diffusion model comprises a pipeline of multiple diffusion models that generate images of increasing resolution, beginning with a standard diffusion model at the lowest resolution, followed by one or more super-resolution diffusion models that successively upsample the image and add higher resolution details. We find that the sample quality of a cascading pipeline relies crucially on conditioning augmentation, our proposed method of data augmentation of the lower resolution conditioning inputs to the super-resolution models. Our experiments show that conditioning augmentation prevents compounding error during sampling in a cascaded model, helping us to train cascading pipelines achieving FID scores of 1.48 at 64x64, 3.52 at 128x128 and 4.88 at 256x256 resolutions, outperforming BigGAN-deep, and classification accuracy scores of 63.02% (top-1) and 84.06% (top-5) at 256x256, outperforming VQ-VAE-2.

Chitwan Saharia, Jonathan Ho, William Chan et al.

We present SR3, an approach to image Super-Resolution via Repeated Refinement. SR3 adapts denoising diffusion probabilistic models to conditional image generation and performs super-resolution through a stochastic denoising process. Inference starts with pure Gaussian noise and iteratively refines the noisy output using a U-Net model trained on denoising at various noise levels. SR3 exhibits strong performance on super-resolution tasks at different magnification factors, on faces and natural images. We conduct human evaluation on a standard 8X face super-resolution task on CelebA-HQ, comparing with SOTA GAN methods. SR3 achieves a fool rate close to 50%, suggesting photo-realistic outputs, while GANs do not exceed a fool rate of 34%. We further show the effectiveness of SR3 in cascaded image generation, where generative models are chained with super-resolution models, yielding a competitive FID score of 11.3 on ImageNet.

Fartash Faghri, Sven Gowal, Cristina Vasconcelos et al.

We demonstrate that the choice of optimizer, neural network architecture, and regularizer significantly affect the adversarial robustness of linear neural networks, providing guarantees without the need for adversarial training. To this end, we revisit a known result linking maximally robust classifiers and minimum norm solutions, and combine it with recent results on the implicit bias of optimizers. First, we show that, under certain conditions, it is possible to achieve both perfect standard accuracy and a certain degree of robustness, simply by training an overparametrized model using the implicit bias of the optimization. In that regime, there is a direct relationship between the type of the optimizer and the attack to which the model is robust. To the best of our knowledge, this work is the first to study the impact of optimization methods such as sign gradient descent and proximal methods on adversarial robustness. Second, we characterize the robustness of linear convolutional models, showing that they resist attacks subject to a constraint on the Fourier-$\ell_\infty$ norm. To illustrate these findings we design a novel Fourier-$\ell_\infty$ attack that finds adversarial examples with controllable frequencies. We evaluate Fourier-$\ell_\infty$ robustness of adversarially-trained deep CIFAR-10 models from the standard RobustBench benchmark and visualize adversarial perturbations.

Sara Sabour, Andrea Tagliasacchi, Soroosh Yazdani et al.

Capsule networks aim to parse images into a hierarchy of objects, parts and relations. While promising, they remain limited by an inability to learn effective low level part descriptions. To address this issue we propose a way to learn primary capsule encoders that detect atomic parts from a single image. During training we exploit motion as a powerful perceptual cue for part definition, with an expressive decoder for part generation within a layered image model with occlusion. Experiments demonstrate robust part discovery in the presence of multiple objects, cluttered backgrounds, and occlusion. The part decoder infers the underlying shape masks, effectively filling in occluded regions of the detected shapes. We evaluate FlowCapsules on unsupervised part segmentation and unsupervised image classification.

Fartash Faghri, David Duvenaud, David J. Fleet et al.

The impact of gradient noise on training deep models is widely acknowledged but not well understood. In this context, we study the distribution of gradients during training. We introduce a method, Gradient Clustering, to minimize the variance of average mini-batch gradient with stratified sampling. We prove that the variance of average mini-batch gradient is minimized if the elements are sampled from a weighted clustering in the gradient space. We measure the gradient variance on common deep learning benchmarks and observe that, contrary to common assumptions, gradient variance increases during training, and smaller learning rates coincide with higher variance. In addition, we introduce normalized gradient variance as a statistic that better correlates with the speed of convergence compared to gradient variance.

Sajad Norouzi, David J. Fleet, Mohammad Norouzi

We introduce Exemplar VAEs, a family of generative models that bridge the gap between parametric and non-parametric, exemplar based generative models. Exemplar VAE is a variant of VAE with a non-parametric prior in the latent space based on a Parzen window estimator. To sample from it, one first draws a random exemplar from a training set, then stochastically transforms that exemplar into a latent code and a new observation. We propose retrieval augmented training (RAT) as a way to speed up Exemplar VAE training by using approximate nearest neighbor search in the latent space to define a lower bound on log marginal likelihood. To enhance generalization, model parameters are learned using exemplar leave-one-out and subsampling. Experiments demonstrate the effectiveness of Exemplar VAEs on density estimation and representation learning. Importantly, generative data augmentation using Exemplar VAEs on permutation invariant MNIST and Fashion MNIST reduces classification error from 1.17% to 0.69% and from 8.56% to 8.16%.

Micha Livne, Kevin Swersky, David J. Fleet

SentenceMIM is a probabilistic auto-encoder for language data, trained with Mutual Information Machine (MIM) learning to provide a fixed length representation of variable length language observations (i.e., similar to VAE). Previous attempts to learn VAEs for language data faced challenges due to posterior collapse. MIM learning encourages high mutual information between observations and latent variables, and is robust against posterior collapse. As such, it learns informative representations whose dimension can be an order of magnitude higher than existing language VAEs. Importantly, the SentenceMIM loss has no hyper-parameters, simplifying optimization. We compare sentenceMIM with VAE, and AE on multiple datasets. SentenceMIM yields excellent reconstruction, comparable to AEs, with a rich structured latent space, comparable to VAEs. The structured latent representation is demonstrated with interpolation between sentences of different lengths. We demonstrate the versatility of sentenceMIM by utilizing a trained model for question-answering and transfer learning, without fine-tuning, outperforming VAE and AE with similar architectures.

Micha Livne, Kevin Swersky, David J. Fleet

We introduce the Mutual Information Machine (MIM), a probabilistic auto-encoder for learning joint distributions over observations and latent variables. MIM reflects three design principles: 1) low divergence, to encourage the encoder and decoder to learn consistent factorizations of the same underlying distribution; 2) high mutual information, to encourage an informative relation between data and latent variables; and 3) low marginal entropy, or compression, which tends to encourage clustered latent representations. We show that a combination of the Jensen-Shannon divergence and the joint entropy of the encoding and decoding distributions satisfies these criteria, and admits a tractable cross-entropy bound that can be optimized directly with Monte Carlo and stochastic gradient descent. We contrast MIM learning with maximum likelihood and VAEs. Experiments show that MIM learns representations with high mutual information, consistent encoding and decoding distributions, effective latent clustering, and data log likelihood comparable to VAE, while avoiding posterior collapse.

Micha Livne, Kevin Swersky, David J. Fleet

We introduce the Mutual Information Machine (MIM), a novel formulation of representation learning, using a joint distribution over the observations and latent state in an encoder/decoder framework. Our key principles are symmetry and mutual information, where symmetry encourages the encoder and decoder to learn different factorizations of the same underlying distribution, and mutual information, to encourage the learning of useful representations for downstream tasks. Our starting point is the symmetric Jensen-Shannon divergence between the encoding and decoding joint distributions, plus a mutual information encouraging regularizer. We show that this can be bounded by a tractable cross entropy loss function between the true model and a parameterized approximation, and relate this to the maximum likelihood framework. We also relate MIM to variational autoencoders (VAEs) and demonstrate that MIM is capable of learning symmetric factorizations, with high mutual information that avoids posterior collapse.

Karen Ullrich, Rianne van den Berg, Marcus Brubaker et al.

Scientific imaging techniques such as optical and electron microscopy and computed tomography (CT) scanning are used to study the 3D structure of an object through 2D observations. These observations are related to the original 3D object through orthogonal integral projections. For common 3D reconstruction algorithms, computational efficiency requires the modeling of the 3D structures to take place in Fourier space by applying the Fourier slice theorem. At present, it is unclear how to differentiate through the projection operator, and hence current learning algorithms can not rely on gradient based methods to optimize 3D structure models. In this paper we show how back-propagation through the projection operator in Fourier space can be achieved. We demonstrate the validity of the approach with experiments on 3D reconstruction of proteins. We further extend our approach to learning probabilistic models of 3D objects. This allows us to predict regions of low sampling rates or estimate noise. A higher sample efficiency can be reached by utilizing the learned uncertainties of the 3D structure as an unsupervised estimate of the model fit. Finally, we demonstrate how the reconstruction algorithm can be extended with an amortized inference scheme on unknown attributes such as object pose. Through empirical studies we show that joint inference of the 3D structure and the object pose becomes more difficult when the ground truth object contains more symmetries. Due to the presence of for instance (approximate) rotational symmetries, the pose estimation can easily get stuck in local optima, inhibiting a fine-grained high-quality estimate of the 3D structure.

Micha Livne, David Fleet

We formulate a new class of conditional generative models based on probability flows. Trained with maximum likelihood, it provides efficient inference and sampling from class-conditionals or the joint distribution, and does not require a priori knowledge of the number of classes or the relationships between classes. This allows one to train generative models from multiple, heterogeneous datasets, while retaining strong prior models over subsets of the data (e.g., from a single dataset, class label, or attribute). In this paper, in addition to end-to-end learning, we show how one can learn a single model from multiple datasets with a relatively weak Glow architecture, and then extend it by conditioning on different knowledge types (e.g., a single dataset). This yields log likelihood comparable to state-of-the-art, compelling samples from conditional priors.

Micha Livne, Leonid Sigal, Marcus A. Brubaker et al.

We propose a generative approach to physics-based motion capture. Unlike prior attempts to incorporate physics into tracking that assume the subject and scene geometry are calibrated and known a priori, our approach is automatic and online. This distinction is important since calibration of the environment is often difficult, especially for motions with props, uneven surfaces, or outdoor scenes. The use of physics in this context provides a natural framework to reason about contact and the plausibility of recovered motions. We propose a fast data-driven parametric body model, based on linear-blend skinning, which decouples deformations due to pose, anthropometrics and body shape. Pose (and shape) parameters are estimated using robust ICP optimization with physics-based dynamic priors that incorporate contact. Contact is estimated from torque trajectories and predictions of which contact points were active. To our knowledge, this is the first approach to take physics into account without explicit {\em a priori} knowledge of the environment or body dimensions. We demonstrate effective tracking from a noisy single depth camera, improving on state-of-the-art results quantitatively and producing better qualitative results, reducing visual artifacts like foot-skate and jitter.

Micha Livne, David J. Fleet

We introduce TzK (pronounced "task"), a conditional probability flow-based model that exploits attributes (e.g., style, class membership, or other side information) in order to learn tight conditional prior around manifolds of the target observations. The model is trained via approximated ML, and offers efficient approximation of arbitrary data sample distributions (similar to GAN and flow-based ML), and stable training (similar to VAE and ML), while avoiding variational approximations. TzK exploits meta-data to facilitate a bottleneck, similar to autoencoders, thereby producing a low-dimensional representation. Unlike autoencoders, the bottleneck does not limit model expressiveness, similar to flow-based ML. Supervised, unsupervised, and semi-supervised learning are supported by replacing missing observations with samples from learned priors. We demonstrate TzK by training jointly on MNIST and Omniglot datasets with minimal preprocessing, and weak supervision, with results comparable to state-of-the-art.

Farzaneh Mahdisoltani, Roland Memisevic, David Fleet

We introduce a hierarchical architecture for video understanding that exploits the structure of real world actions by capturing targets at different levels of granularity. We design the model such that it first learns simpler coarse-grained tasks, and then moves on to learn more fine-grained targets. The model is trained with a joint loss on different granularity levels. We demonstrate empirical results on the recent release of Something-Something dataset, which provides a hierarchy of targets, namely coarse-grained action groups, fine-grained action categories, and captions. Experiments suggest that models that exploit targets at different levels of granularity achieve better performance on all levels.

Farzaneh Mahdisoltani, Guillaume Berger, Waseem Gharbieh et al.

We describe a DNN for video classification and captioning, trained end-to-end, with shared features, to solve tasks at different levels of granularity, exploring the link between granularity in a source task and the quality of learned features for transfer learning. For solving the new task domain in transfer learning, we freeze the trained encoder and fine-tune a neural net on the target domain. We train on the Something-Something dataset with over 220, 000 videos, and multiple levels of target granularity, including 50 action groups, 174 fine-grained action categories and captions. Classification and captioning with Something-Something are challenging because of the subtle differences between actions, applied to thousands of different object classes, and the diversity of captions penned by crowd actors. Our model performs better than existing classification baselines for SomethingSomething, with impressive fine-grained results. And it yields a strong baseline on the new Something-Something captioning task. Experiments reveal that training with more fine-grained tasks tends to produce better features for transfer learning.

Azin Asgarian, Ahmed Bilal Ashraf, David Fleet et al.

Active appearance models (AAMs) are a class of generative models that have seen tremendous success in face analysis. However, model learning depends on the availability of detailed annotation of canonical landmark points. As a result, when accurate AAM fitting is required on a different set of variations (expression, pose, identity), a new dataset is collected and annotated. To overcome the need for time consuming data collection and annotation, transfer learning approaches have received recent attention. The goal is to transfer knowledge from previously available datasets (source) to a new dataset (target). We propose a subspace transfer learning method, in which we select a subspace from the source that best describes the target space. We propose a metric to compute the directional similarity between the source eigenvectors and the target subspace. We show an equivalence between this metric and the variance of target data when projected onto source eigenvectors. Using this equivalence, we select a subset of source principal directions that capture the variance in target data. To define our model, we augment the selected source subspace with the target subspace learned from a handful of target examples. In experiments done on six publicly available datasets, we show that our approach outperforms the state of the art in terms of the RMS fitting error as well as the percentage of test examples for which AAM fitting converges to the ground truth.

Fartash Faghri, David J. Fleet, Jamie Ryan Kiros et al.

We present a new technique for learning visual-semantic embeddings for cross-modal retrieval. Inspired by hard negative mining, the use of hard negatives in structured prediction, and ranking loss functions, we introduce a simple change to common loss functions used for multi-modal embeddings. That, combined with fine-tuning and use of augmented data, yields significant gains in retrieval performance. We showcase our approach, VSE++, on MS-COCO and Flickr30K datasets, using ablation studies and comparisons with existing methods. On MS-COCO our approach outperforms state-of-the-art methods by 8.8% in caption retrieval and 11.3% in image retrieval (at R@1).

Yanshuai Cao, David J. Fleet

We introduce a framework for analyzing transductive combination of Gaussian process (GP) experts, where independently trained GP experts are combined in a way that depends on test point location, in order to scale GPs to big data. The framework provides some theoretical justification for the generalized product of GP experts (gPoE-GP) which was previously shown to work well in practice but lacks theoretical basis. Based on the proposed framework, an improvement over gPoE-GP is introduced and empirically validated.