Visual data is used in numerous different scientific work-flows ranging from remote sensing to ecology. As the amount of observation data increases, the challenge is not just to make accurate predictions but also to understand the underlying mechanisms for those predictions. Good interpretation is important in scientific workflows, as it allows for better decision-making by providing insights into the data. This paper introduces an automatic way of obtaining such interpretable-by-design models, by learning programs that interleave neural networks. We propose DiSciPLE (Discovering Scientific Programs using LLMs and Evolution) an evolutionary algorithm that leverages common sense and prior knowledge of large language models (LLMs) to create Python programs explaining visual data. Additionally, we propose two improvements: a program critic and a program simplifier to improve our method further to synthesize good programs. On three different real-world problems, DiSciPLE learns state-of-the-art programs on novel tasks with no prior literature. For example, we can learn programs with 35% lower error than the closest non-interpretable baseline for population density estimation.

It is difficult to anticipate the myriad challenges that a predictive model will encounter once deployed. Common practice entails a reactive, cyclical approach: model deployment, data mining, and retraining. We instead develop a proactive longtail discovery process by imagining additional data during training. In particular, we develop general model-based longtail signals, including a differentiable, single forward pass formulation of epistemic uncertainty that does not impact model parameters or predictive performance but can flag rare or hard inputs. We leverage these signals as guidance to generate additional training data from a latent diffusion model in a process we call Longtail Guidance (LTG). Crucially, we can perform LTG without retraining the diffusion model or the predictive model, and we do not need to expose the predictive model to intermediate diffusion states. Data generated by LTG exhibit semantically meaningful variation, yield significant generalization improvements on numerous image classification benchmarks, and can be analyzed by a VLM to proactively discover, textually explain, and address conceptual gaps in a deployed predictive model.

Underwater robotic manipulation faces significant challenges due to complex fluid dynamics and unstructured environments, causing most manipulation systems to rely heavily on human teleoperation. In this paper, we introduce AquaBot, a fully autonomous manipulation system that combines behavior cloning from human demonstrations with self-learning optimization to improve beyond human teleoperation performance. With extensive real-world experiments, we demonstrate AquaBot's versatility across diverse manipulation tasks, including object grasping, trash sorting, and rescue retrieval. Our real-world experiments show that AquaBot's self-optimized policy outperforms a human operator by 41% in speed. AquaBot represents a promising step towards autonomous and self-improving underwater manipulation systems. We will open-source both hardware and software implementation details.

This work-in-progress proposes an approach that uses a low-cost, smartphone-based system for at-home, inquiry-driven science learning called STEM-Messaging System (SMS). SMS supports real-time, interactive, message-based science activities and is part of a broader project aimed at integrating science into children's daily lives by uncovering the science behind everyday objects via computer vision overlays. We discuss how three pedagogical principles–inquiry-based learning, culturally relevant pedagogies, and modeling-based learning–inform key design features of the system and its curricular activities. We identify tensions that surfaced from pilot studies involving students, parents, and teachers, providing examples of how pedagogical principles and practical applications influence design decisions.

Advancements in 3D Gaussian Splatting have significantly accelerated 3D reconstruction and generation. However, it may require a large number of Gaussians, which creates a substantial memory footprint. This paper introduces GES (Generalized Exponential Splatting), a novel representation that employs Generalized Exponential Function (GEF) to model 3D scenes, requiring far fewer parti-cles to represent a scene and thus significantly outperforming Gaussian Splatting methods in efficiency with a plug-and-play replacement ability for Gaussian-based utilities. GES is validated theoretically and empirically in both principled 1D setup and realistic 3D scenes. It is shown to represent signals with sharp edges more accurately, which are typically challenging for Gaussians due to their inherent low-pass characteristics. Our empirical analysis demonstrates that GEF outperforms Gaussians in fitting natural-occurring signals (e.g. squares, triangles, parabolic signals), thereby reducing the need for extensive splitting operations that increase the memory footprint of Gaussian Splatting. With the aid of a frequency-modulated loss, GES achieves competitive performance in novel-view syn-thesis benchmarks while requiring less than half the memory storage of Gaussian Splatting and increasing the rendering speed by up to 39%. The code is available on the project website https://abdullahamdi.com/ges.

We introduce pix2gestalt, a framework for zero-shot amodal segmentation, which learns to estimate the shape and appearance of whole objects that are only partially visible behind occlusions. By capitalizing on large-scale diffusion models and transferring their representations to this task, we learn a conditional diffusion model for reconstructing whole objects in challenging zero-shot cases, including examples that break natural and physical priors, such as art. As training data, we use a synthetically curated dataset containing occluded objects paired with their whole counterparts. Experiments show that our approach outperforms supervised baselines on established benchmarks. Our model can furthermore be used to significantly improve the performance of existing object recognition and 3D reconstruction methods in the presence of occlusions.

When presented with questions involving visual thinking, humans naturally switch reasoning modalities, often forming mental images or drawing visual aids. Large language models have shown promising results in arithmetic and symbolic reasoning by expressing intermediate reasoning in text as a chain of thought, yet struggle to extend this capability to answer text queries that are easily solved by visual reasoning, even with extensive multimodal pretraining. We introduce a simple method, whiteboard-of-thought prompting, to unlock the visual reasoning capabilities of multimodal large language models across modalities. Whiteboard-of-thought prompting provides multimodal large language models with a metaphorical ‘whiteboard’ to draw out reasoning steps as images, then returns these images back to the model for further processing. We find this can be accomplished with no demonstrations or specialized modules, instead leveraging models’ existing ability to write code with libraries such as Matplotlib and Turtle. This simple approach shows state-of-the-art results on four difficult natural language tasks that involve visual and spatial reasoning. We identify multiple settings where GPT-4o using chain-of-thought fails dramatically, including more than one where it achieves 0% accuracy, while whiteboard-of-thought enables up to 92% accuracy in these same settings. We present a detailed exploration of where the technique succeeds as well as its sources of error.

Large-scale pre-trained vision foundation models, such as CLIP, have become de facto backbones for various vision tasks. However, due to their black-box nature, understanding the underlying rules behind these models’ predictions and controlling model behaviors have remained open challenges. We present INViTE: a framework for INterpreting Vision Transformer’s latent tokens with Text Explanations. Given a latent token, INViTE retains its semantic information to the final layer using transformer’s local operations and retrieves the closest text for explanation. INViTE enables understanding of model visual reasoning procedure without needing additional model training or data collection. Based on the obtained interpretations, INViTE allows for model editing that controls model reasoning behaviors and improves model robustness against biases and spurious correlations. Our code is available at https://github.com/tonychenxyz/vit-interpret.

We introduce a method to train vision-language models for remote-sensing images without using any textual annotations. Our key insight is to use co-located internet imagery taken on the ground as an intermediary for connecting remote-sensing images and language. Specifically, we train an image encoder for remote sensing images to align with the image encoder of CLIP using a large amount of paired internet and satellite images. Our unsupervised approach enables the training of a first-of-its-kind large scale VLM for remote sensing images at two different resolutions. We show that these VLMs enable zero-shot, open-vocabulary image classification, retrieval, segmentation and visual question answering for satellite images. On each of these tasks, our VLM trained without textual annotations outperforms existing VLMs trained with supervision, with gains of up to 20\% for classification and 80\% for segmentation.

We find that large language models (LLMs) are more likely to modify human-written text than AI-generated text when tasked with rewriting. This tendency arises because LLMs often perceive AI-generated text as high-quality, leading to fewer modifications. We introduce a method to detect AI-generated content by prompting LLMs to rewrite text and calculating the editing distance of the output. We dubbed our geneRative AI Detection viA Rewriting method Raidar. Raidar significantly improves the F1 detection scores of existing AI content detection models -- both academic and commercial -- across various domains, including News, creative writing, student essays, code, Yelp reviews, and arXiv papers, with gains of up to 29 points. Operating solely on word symbols without high-dimensional features, our method is compatible with black box LLMs, and is inherently robust on new content. Our results illustrate the unique imprint of machine-generated text through the lens of the machines themselves.

Synthesizing novel 3D models that resemble the input example as long been pursued by graphics artists and machine learning researchers. In this paper, we present Sin3DM, a diffusion model that learns the internal patch distribution from a single 3D textured shape and generates high-quality variations with fine geometry and texture details. Training a diffusion model directly in 3D would induce large memory and computational cost. Therefore, we first compress the input into a lower-dimensional latent space and then train a diffusion model on it. Specifically, we encode the input 3D textured shape into triplane feature maps that represent the signed distance and texture fields of the input. The denoising network of our diffusion model has a limited receptive field to avoid overfitting, and uses triplane-aware 2D convolution blocks to improve the result quality. Aside from randomly generating new samples, our model also facilitates applications such as retargeting, outpainting and local editing. Through extensive qualitative and quantitative evaluation, we show that our method outperforms prior methods in generation quality of 3D shapes.

How do large language models (LLMs) obtain their answers? The ability to explain and control an LLM’s reasoning process is key for reliability, transparency, and future model developments. We propose SelfIE (Self-Interpretation of Embeddings), a framework that enables LLMs to interpret their own embeddings in natural language by leveraging their ability to respond to inquiries about a given passage. Capable of interpreting open-world concepts in the hidden embeddings, SelfIE reveals LLM internal reasoning in cases such as making ethical decisions, internalizing prompt injection, and recalling harmful knowledge. SelfIE’s text descriptions on hidden embeddings open avenues to control LLM reasoning. We propose Supervised Control, which allows editing open-ended concepts while only requiring gradient computation of individual layer. We extend RLHF to hidden embeddings and propose Reinforcement Control that erases harmful knowledge in LLM without supervision targets.

We present MIRACLE, a system for online, interpretable visual concept and video action recognition. Through a chat interface, users query the recognition system with an uploaded image or video. For images, MIRACLE returns concept predictions from its structured knowledge base, justifying its predictions with heatmaps and natural language-based attribute detections. For videos, MIRACLE predicts an action and justifies its prediction with time varying entity-entity relations. With its ability to learn new concepts in an online, few-shot manner and its support of dynamic changes to its knowledge base, MIRACLE represents a step forward in interpretable multimodal learning systems.

Tracking objects with persistence in cluttered and dynamic environments remains a difficult challenge for computer vision systems. In this paper, we introduce TCOW, a new benchmark and model for visual tracking through heavy occlusion and containment. We set up a task where the goal is to, given a video sequence, segment both the projected extent of the target object, as well as the surrounding container or occluder whenever one exists. To study this task, we create a mixture of synthetic and annotated real datasets to support both supervised learning and structured evaluation of model performance under various forms of task variation, such as moving or nested containment. We evaluate two recent transformer-based video models and find that while they can be surprisingly capable of tracking targets under certain settings of task variation, there remains a considerable performance gap before we can claim a tracking model to have acquired a true notion of object permanence.

3D reconstruction is a fundamental problem in computer vision, and the task is especially challenging when the object to reconstruct is partially or fully occluded. We introduce a method that uses the shadows cast by an unobserved object in order to infer the possible 3D volumes under occlusion. We create a differentiable image formation model that allows us to jointly infer the 3D shape of an object, its pose, and the position of a light source. Since the approach is end-to-end differentiable, we are able to integrate learned priors of object geometry in order to generate realistic 3D shapes of different object categories. Experiments and visualizations show that the method is able to generate multiple possible solutions that are consistent with the observation of the shadow. Our approach works even when the position of the light source and object pose are both unknown. Our approach is also robust to real-world images where ground-truth shadow mask is unknown.

The relatively hot temperature of the human body causes people to turn into long-wave infrared light sources. Since this emitted light has a larger wavelength than visible light, many surfaces in typical scenes act as infrared mirrors with strong specular reflections. We exploit the thermal reflections of a person onto objects in order to locate their position and reconstruct their pose, even if they are not visible to a normal camera. We propose an analysis-by-synthesis framework that jointly models the objects, people, and their thermal reflections, which combines generative models with differentiable rendering of reflections. Quantitative and qualitative experiments show our approach works in highly challenging cases, such as with curved mirrors or when the person is completely unseen by a normal camera.

Many visual recognition models are evaluated only on their classification accuracy, a metric for which they obtain strong performance. In this paper, we investigate whether computer vision models can also provide correct rationales for their predictions. We propose a “doubly right” object recognition benchmark, where the metric requires the model to simultaneously produce both the right labels as well as the right rationales. We find that state-of-the-art visual models, such as CLIP, often provide incorrect rationales for their categorical predictions. However, by transferring the rationales from language models into visual representations through a tailored dataset, we show that we can learn a “why prompt,” which adapts large visual representations to produce correct rationales. Visualizations and empirical experiments show that our prompts significantly improve performance on doubly right object recognition, in addition to zero-shot transfer to unseen tasks and datasets.

Synthesizing 3D human avatars interacting realistically with a scene is an important problem with applications in AR/VR, video games, and robotics. Towards this goal, we address the task of generating a virtual human – hands and full body – grasping everyday objects. Existing methods approach this problem by collecting a 3D dataset of humans interacting with objects and training on this data. However, 1) these methods do not generalize to different object positions and orientations or to the presence of furniture in the scene, and 2) the diversity of their generated full-body poses is very limited. In this work, we address all the above challenges to generate realistic, diverse full-body grasps in everyday scenes without requiring any 3D full-body grasping data. Our key insight is to leverage the existence of both full-body pose and hand-grasping priors, composing them using 3D geometrical constraints to obtain full-body grasps. We empirically validate that these constraints can generate a variety of feasible human grasps that are superior to baselines both quantitatively and qualitatively. See our webpage for more details: flex.cs.columbia.edu.

We present SHIFT3D, a differentiable pipeline for generating 3D shapes that are structurally plausible yet challenging to 3D object detectors. In safety-critical applications like autonomous driving, discovering such novel challenging objects can offer insight into unknown vulnerabilities of 3D detectors. By representing objects with a signed distanced function (SDF), we show that gradient error signals allow us to smoothly deform the shape or pose of a 3D object in order to confuse a downstream 3D detector. Importantly, the objects generated by SHIFT3D physically differ from the baseline object yet retain a semantically recognizable shape. Our approach provides interpretable failure modes for modern 3D object detectors, and can aid in preemptive discovery of potential safety risks within 3D perception systems before these risks become critical failures.

Human motion is created by, and constrained by, our muscles. We take a first step at building computer vision methods that represent the internal muscle activity that causes motion. We present a new dataset, Muscles in Action (MIA), to learn to incorporate muscle activity into human motion representations. The dataset consists of 12.5 hours of synchronized video and surface electromyography (sEMG) data of 10 subjects performing various exercises. Using this dataset, we learn a bidirectional representation that predicts muscle activation from video, and conversely, reconstructs motion from muscle activation. We evaluate our model on in-distribution subjects and exercises, as well as on out-of-distribution subjects and exercises. We demonstrate how advances in modeling both modalities jointly can serve as conditioning for muscularly consistent motion generation. Putting muscles into computer vision systems will enable richer models of virtual humans, with applications in sports, fitness, and AR/VR.

Many machine learning methods operate by inverting a neural network at inference time, which has become a popular technique for solving inverse problems in computer vision, robotics, and graphics. However, these methods often involve gradient descent through a highly non-convex loss landscape, causing the optimization process to be unstable and slow. We introduce a method that learns a loss landscape where gradient descent is efficient, bringing massive improvement and acceleration to the inversion process. We demonstrate this advantage on a number of methods for both generative and discriminative tasks, including GAN inversion, adversarial defense, and 3D human pose reconstruction.

We introduce Zero-1-to-3, a framework for changing the camera viewpoint of an object given just a single RGB image. To perform novel view synthesis in this under-constrained setting, we capitalize on the geometric priors that large-scale diffusion models learn about natural images. Our conditional diffusion model uses a synthetic dataset to learn controls of the relative camera viewpoint, which allow new images to be generated of the same object under a specified camera transformation. Even though it is trained on a synthetic dataset, our model retains a strong zero-shot generalization ability to out-of-distribution datasets as well as in-the-wild images, including impressionist paintings. Our viewpoint-conditioned diffusion approach can further be used for the task of 3D reconstruction from a single image. Qualitative and quantitative experiments show that our method significantly outperforms state-of-the-art single-view 3D reconstruction and novel view synthesis models by leveraging Internet-scale pre-training.

Modeling the mechanics of fluid in complex scenes is vital to applications in design, graphics, and robotics. Learning-based methods provide fast and differentiable fluid simulators, however most prior work is unable to accurately model how fluids interact with genuinely novel surfaces not seen during training. We introduce SurfsUp, a framework that represents objects implicitly using signed distance functions (SDFs), rather than an explicit representation of meshes or particles. This continuous representation of geometry enables more accurate simulation of fluid-object interactions over long time periods while simultaneously making computation more efficient. Moreover, SurfsUp trained on simple shape primitives generalizes considerably out-of-distribution, even to complex real-world scenes and objects. Finally, we show we can invert our model to design simple objects to manipulate fluid flow.

Answering visual queries is a complex task that requires both visual processing and reasoning. End-to-end models, the dominant approach for this task, do not explicitly differentiate between the two, limiting interpretability and generalization. Learning modular programs presents a promising alternative, but has proven challenging due to the difficulty of learning both the programs and modules simultaneously. We introduce ${\color{green}{\text{ViperGPT}}}$, a framework that leverages code-generation models to compose vision-and-language models into subroutines to produce a result for any query. ${\color{green}{\text{ViperGPT}}}$ utilizes a provided API to access the available modules, and composes them by generating Python code that is later executed. This simple approach requires no further training, and achieves state-of-the-art results across various complex visual tasks.

Deep networks for computer vision are not reliable when they encounter adversarial examples. In this paper, we introduce a framework that uses the dense intrinsic constraints in natural images to robustify inference. By introducing constraints at inference time, we can shift the burden of robustness from training to testing, thereby allowing the model to dynamically adjust to each individual image's unique and potentially novel characteristics at inference time. Our theoretical results show the importance of having dense constraints at inference time. In contrast to existing single-constraint methods, we propose to use equivariance, which naturally allows dense constraints at a fine-grained level in the feature space. Our empirical experiments show that restoring feature equivariance at inference time defends against worst-case adversarial perturbations. The method obtains improved adversarial robustness on four datasets (ImageNet, Cityscapes, PASCAL VOC, and MS-COCO) on image recognition, semantic segmentation, and instance segmentation tasks.

Natural language processing and 2D vision models have attained remarkable proficiency on many tasks primarily by escalating the scale of training data. However, 3D vision tasks have not seen the same progress, in part due to the challenges of acquiring high-quality 3D data. In this work, we present Objaverse-XL, a dataset of over 10 million 3D objects. Our compilation comprises deduplicated 3D objects from a diverse set of sources, including manually designed objects, photogrammetry scans of landmarks and everyday items, and professional scans of historic and antique artifacts. Representing the largest scale and diversity in the realm of 3D datasets, Objaverse-XL enables significant new possibilities for 3D vision. Our experiments demonstrate the vast improvements enabled with the scale provided by Objaverse-XL. We show that by training Zero123 on novel view synthesis, utilizing over 100 million multi-view rendered images, we achieve strong zero-shot generalization abilities. We hope that releasing Objaverse-XL will enable further innovations in the field of 3D vision at scale.

Modern climate projections lack adequate spatial and temporal resolution due to computational constraints. A consequence is inaccurate and imprecise predictions of critical processes such as storms. Hybrid methods that combine physics with machine learning (ML) have introduced a new generation of higher fidelity climate simulators that can sidestep Moore's Law by outsourcing compute-hungry, short, high-resolution simulations to ML emulators. However, this hybrid ML-physics simulation approach requires domain-specific treatment and has been inaccessible to ML experts because of lack of training data and relevant, easy-to-use workflows. We present ClimSim, the largest-ever dataset designed for hybrid ML-physics research. It comprises multi-scale climate simulations, developed by a consortium of climate scientists and ML researchers. It consists of 5.7 billion pairs of multivariate input and output vectors that isolate the influence of locally-nested, high-resolution, high-fidelity physics on a host climate simulator's macro-scale physical state.The dataset is global in coverage, spans multiple years at high sampling frequency, and is designed such that resulting emulators are compatible with downstream coupling into operational climate simulators. We implement a range of deterministic and stochastic regression baselines to highlight the ML challenges and their scoring. The data (https://huggingface.co/datasets/LEAP/ClimSim_high-res) and code (https://leap-stc.github.io/ClimSim) are released openly to support the development of hybrid ML-physics and high-fidelity climate simulations for the benefit of science and society.

Images are often more significant than only the pixels to human eyes, as we can infer, associate, and reason with contextual information from other sources to establish a more complete picture. For example, in Figure This reasoning could provide the time and place the image was taken, which will help us in subsequent tasks, such as automatic storyline construction, correction of image source in intended effect photographs, and upper-stream processing such as image clustering for certain location or time.

For computer vision systems to operate in dynamic situations, they need to be able to represent and reason about object permanence. We introduce a framework for learning to estimate 4D visual representations from monocular RGB-D video, which is able to persist objects, even once they become obstructed by occlusions. Unlike traditional video representations, we encode point clouds into a continuous representation, which permits the model to attend across the spatiotemporal context to resolve occlusions. On two large video datasets that we release along with this paper, our experiments show that the representation is able to successfully reveal occlusions for several tasks, without any architectural changes. Visualizations show that the attention mechanism automatically learns to follow occluded objects. Since our approach can be trained end-to-end and is easily adaptable, we believe it will be useful for handling occlusions in many video understanding tasks. Data, code, and models are available at occ1usions. cs. co1umbia. edu.

Visual representations underlie object recognition tasks, but they often contain both robust and non-robust features. Our main observation is that image classifiers may perform poorly on out-of-distribution samples because spurious correlations between non-robust features and labels can be changed in a new environment. By analyzing procedures for out-of-distribution generalization with a causal graph, we show that standard classifiers fail because the association between images and labels is not transportable across settings. However, we then show that the causal effect, which severs all sources of confounding, remains invariant across domains. This motivates us to develop an algorithm to estimate the causal effect for image classification, which is transportable (i.e., invariant) across source and target environments. Without observing additional variables, we show that we can derive an estimand for the causal effect under empirical assumptions using representations in deep models as proxies. Theoretical analysis, empirical results, and visualizations show that our approach captures causal invariances and improves overall generalization.

Our lives can be seen as a complex weaving of activities; we switch from one activity to another, to maximise our achievements or in reaction to demands placed upon us. Observing a video of unscripted daily activities, we parse the video into its constituent activity threads through a process we call unweaving. To accomplish this, we introduce a video representation explicitly capturing activity threads called a thread bank, along with a neural controller capable of detecting goal changes and resuming of past activities, together forming UnweaveNet. We train and evaluate UnweaveNet on sequences from the unscripted egocentric dataset EPIC-KITCHENS. We propose and showcase the efficacy of pretraining UnweaveNet in a self-supervised manner.

Machine translation between many languages at once is highly challenging, since training with ground truth re-quires supervision between all language pairs, which is dif-ficult to obtain. Our key insight is that, while languages may vary drastically, the underlying visual appearance of the world remains consistent. We introduce a method that uses visual observations to bridge the gap between languages, rather than relying on parallel corpora or topo-logical properties of the representations. We train a model that aligns segments of text from different languages if and only if the images associated with them are similar and each image in turn is well-aligned with its textual description. We train our model from scratch on a new dataset of text in over fifty languages with accompanying images. Experiments show that our method outperforms previous work on unsupervised word and sentence translation using retrieval. Code, models and data are available on globetrotter.cs.columbia.edu

We present an approach for recommending a music track for a given video, and vice versa, based on both their temporal alignment and their correspondence at an artistic level. We propose a self-supervised approach that learns this correspondence directly from data, without any need of human annotations. In order to capture the high-level concepts that are required to solve the task, we propose modeling the long-term temporal context of both the video and the music signals, using Transformer networks for each modality. Experiments show that this approach strongly outperforms alternatives that do not exploit the temporal context. The combination of our contributions improve retrieval accuracy up to 10× over prior state of the art. This strong improvement allows us to introduce a wide range of analyses and applications. For instance, we can condition music retrieval based on visually defined attributes.

We introduce a framework for navigating through cluttered environments by connecting multiple cameras together while simultanously preserving privacy. Occlusions and obstacles in large environments are often challenging situations for navigation agents because the environment is not fully observable from a single camera view. Given multiple camera views of an environment, our approach learns to produce a multiview scene representation that can only be used for navigation, provably preventing one party from inferring anything beyond the output task. On a new navigation dataset that we will publicly release, experiments show that private multiparty representations allow navigation through complex scenes and around obstacles while jointly preserving privacy. Our approach scales to an arbitrary number of camera viewpoints. We believe developing visual representations that preserve privacy is increasingly important for many applications such as navigation.

We introduce a representation learning framework for spatial trajectories. We represent partial observations of trajectories as probability distributions in a learned latent space, which characterize the uncertainty about unobserved parts of the trajectory. Our framework allows us to obtain samples from a trajectory for any continuous point in time—both interpolating and extrapolating. Our flexible approach supports directly modifying specific attributes of a trajectory, such as its pace, as well as combining different partial observations into single representations. Experiments show our method's superiority over baselines in prediction tasks.

Variational autoencoders (VAEs) suffer from posterior collapse, where the powerful neural networks used for modeling and inference optimize the objective without meaningfully using the latent representation. We introduce inference critics that detect and incentivize against posterior collapse by requiring correspondence between latent variables and the observations. By connecting the critic’s objective to the literature in self-supervised contrastive representation learning, we show both theoretically and empirically that optimizing inference critics increases the mutual information between observations and latents, mitigating posterior collapse. This approach is straightforward to implement and requires significantly less training time than prior methods, yet obtains competitive results on three established datasets. Overall, the approach lays the foundation to bridge the previously disconnected frameworks of contrastive learning and probabilistic modeling with variational autoencoders, underscoring the benefits both communities may find at their intersection.

Managing inputs that are novel, unknown, or out-of-distribution is critical as an agent moves from the lab to the open world. Novelty-related problems include being tolerant to novel perturbations of the normal input, detecting when the input includes novel items, and adapting to novel inputs. While significant research has been undertaken in these areas, a noticeable gap exists in the lack of a formalized definition of novelty that transcends problem domains. As a team of researchers spanning multiple research groups and different domains, we have seen, first hand, the difficulties that arise from ill-specified novelty problems, as well as inconsistent definitions and terminology. Therefore, we present the first unified framework for formal theories of novelty and use the framework to formally define a family of novelty types. Our framework can be applied across a wide range of domains, from symbolic AI to reinforcement learning, and beyond to open world image recognition. Thus, it can be used to help kick-start new research efforts and accelerate ongoing work on these important novelty-related problems.

We introduce a framework that predicts the goals behind observable human action in video. Motivated by evidence in developmental psychology, we leverage video of unintentional action to learn video representations of goals without direct supervision. Our approach models videos as contextual trajectories that represent both low-level motion and high-level action features. Experiments and visualizations show our trained model is able to predict the underlying goals in video of unintentional action. We also propose a method to "automatically correct" unintentional action by leveraging gradient signals of our model to adjust latent trajectories. Although the model is trained with minimal supervision, it is competitive with or outperforms baselines trained on large (supervised) datasets of successfully executed goals, showing that observing unintentional action is crucial to learning about goals in video.

We introduce a framework for learning robust visual representations that generalize to new viewpoints, backgrounds, and scene contexts. Discriminative models often learn naturally occurring spurious correlations, which cause them to fail on images outside of the training distribution. In this paper, we show that we can steer generative models to manufacture interventions on features caused by confounding factors. Experiments, visualizations, and theoretical results show this method learns robust representations more consistent with the underlying causal relationships. Our approach improves performance on multiple datasets demanding out-of-distribution generalization, and we demonstrate state-of-the-art performance generalizing from ImageNet to ObjectNet dataset.

We introduce a framework for learning from unlabeled video what is predictable in the future. Instead of committing up front to features to predict, our approach learns from data which features are predictable. Based on the observation that hyperbolic geometry naturally and compactly encodes hierarchical structure, we propose a predictive model in hyperbolic space. When the model is most confident, it will predict at a concrete level of the hierarchy, but when the model is not confident, it learns to automatically select a higher level of abstraction. Experiments on two established datasets show the key role of hierarchical representations for action prediction. Although our representation is trained with unlabeled video, visualizations show that action hierarchies emerge in the representation.

The elementary operation of cropping underpins nearly every computer vision system, ranging from data augmentation and translation invariance to computational photography and representation learning. This paper investigates the subtle traces introduced by this operation. For example, despite refinements to camera optics, lenses will leave behind certain clues, notably chromatic aberration and vignetting. Photographers also leave behind other clues relating to image aesthetics and scene composition. We study how to detect these traces, and investigate the impact that cropping has on the image distribution. While our aim is to dissect the fundamental impact of spatial crops, there are also a number of practical implications to our work, such as revealing faulty photojournalism and equipping neural network researchers with a better understanding of shortcut learning. Code is available at https://github.com/basilevh/dissecting-image-crops.

We find that images contain intrinsic structure that enables the reversal of many adversarial attacks. Attack vectors cause not only image classifiers to fail, but also collaterally disrupt incidental structure in the image. We demonstrate that modifying the attacked image to restore the natural structure will reverse many types of attacks, providing a defense. Experiments demonstrate significantly improved robustness for several state-of-the-art models across the CIFAR-10, CIFAR-100, SVHN, and ImageNet datasets. Our results show that our defense is still effective even if the attacker is aware of the defense mechanism. Since our defense is deployed during inference instead of training, it is compatible with pre-trained networks as well as most other defenses. Our results suggest deep networks are vulnerable to adversarial examples partly because their representations do not enforce the natural structure of images.

From just a short glance at a video, we can often tell whether a person's action is intentional or not. Can we train a model to recognize this? We introduce a dataset of in-the-wild videos of unintentional action, as well as a suite of tasks for recognizing, localizing, and anticipating its onset. We train a supervised neural network as a baseline and analyze its performance compared to human consistency on the tasks. We also investigate self-supervised representations that leverage natural signals in our dataset, and show the effectiveness of an approach that uses the intrinsic speed of video to perform competitively with highly-supervised pretraining. However, a significant gap between machine and human performance remains.

We introduce a deep learning model for speech denoising, a long-standing challenge in audio analysis arising in numerous applications. Our approach is based on a key observation about human speech: there is often a short pause between each sentence or word. In a recorded speech signal, those pauses introduce a series of time periods during which only noise is present. We leverage these incidental silent intervals to learn a model for automatic speech denoising given only mono-channel audio. Detected silent intervals over time expose not just pure noise but its time-varying features, allowing the model to learn noise dynamics and suppress it from the speech signal. Experiments on multiple datasets confirm the pivotal role of silent interval detection for speech denoising, and our method outperforms several state-of-the-art denoising methods, including those that accept only audio input (like ours) and those that denoise based on audiovisual input (and hence require more information). We also show that our method enjoys excellent generalization properties, such as denoising spoken languages not seen during training.

We present the Moments in Time Dataset, a large-scale human-annotated collection of one million short videos corresponding to dynamic events unfolding within three seconds. Modeling the spatial-audio-temporal dynamics even for actions occurring in 3 second videos poses many challenges: meaningful events do not include only people, but also objects, animals, and natural phenomena; visual and auditory events can be symmetrical in time ("opening" is "closing" in reverse), and either transient or sustained. We describe the annotation process of our dataset (each video is tagged with one action or activity label among 339 different classes), analyze its scale and diversity in comparison to other large-scale video datasets for action recognition, and report results of several baseline models addressing separately, and jointly, three modalities: spatial, temporal and auditory. The Moments in Time dataset, designed to have a large coverage and diversity of events in both visual and auditory modalities, can serve as a new challenge to develop models that scale to the level of complexity and abstract reasoning that a human processes on a daily basis.

We address the problem of phrase grounding by learning a multi-level common semantic space shared by the textual and visual modalities. We exploit multiple levels of feature maps of a Deep Convolutional Neural Network, as well as contextualized word and sentence embeddings extracted from a character-based language model. Following dedicated non-linear mappings for visual features at each level, word, and sentence embeddings, we obtain multiple instantiations of our common semantic space in which comparisons between any target text and the visual content is performed with cosine similarity. We guide the model by a multi-level multimodal attention mechanism which outputs attended visual features at each level. The best level is chosen to be compared with text content for maximizing the pertinence scores of image-sentence pairs of the ground truth. Experiments conducted on three publicly available datasets show significant performance gains (20%-60% relative) over the state-of-the-art in phrase localization and set a new performance record on those datasets. We provide a detailed ablation study to show the contribution of each element of our approach and release our code on GitHub.

This paper focuses on multi-person action forecasting in videos. More precisely, given a history of H previous frames, the goal is to detect actors and to predict their future actions for the next T frames. Our approach jointly models temporal and spatial interactions among different actors by constructing a recurrent graph, using actor proposals obtained with Faster R-CNN as nodes. Our method learns to select a subset of discriminative relations without requiring explicit supervision, thus enabling us to tackle challenging visual data. We refer to our model as Discriminative Relational Recurrent Network (DRRN). Evaluation of action prediction on AVA demonstrates the effectiveness of our proposed method compared to simpler baselines. Furthermore, we significantly improve performance on the task of early action classification on J-HMDB, from the previous SOTA of 48% to 60%.

Self-supervised learning has become increasingly important to leverage the abundance of unlabeled data available on platforms like YouTube. Whereas most existing approaches learn low-level representations, we propose a joint visual-linguistic model to learn high-level features without any explicit supervision. In particular, inspired by its recent success in language modeling, we build upon the BERT model to learn bidirectional joint distributions over sequences of visual and linguistic tokens, derived from vector quantization of video data and off-the-shelf speech recognition outputs, respectively. We use VideoBERT in numerous tasks, including action classification and video captioning. We show that it can be applied directly to open-vocabulary classification, and confirm that large amounts of training data and cross-modal information are critical to performance. Furthermore, we outperform the state-of-the-art on video captioning, and quantitative results verify that the model learns high-level semantic features.

Deep networks are well-known to be fragile to adversarial attacks. We conduct an empirical analysis of deep representations under the state-of-the-art attack method called PGD, and find that the attack causes the internal representation to shift closer to the ``false'' class. Motivated by this observation, we propose to regularize the representation space under attack with metric learning to produce more robust classifiers. By carefully sampling examples for metric learning, our learned representation not only increases robustness, but also detects previously unseen adversarial samples. Quantitative experiments show improvement of robustness accuracy by up to 4% and detection efficiency by up to 6% according to Area Under Curve score over prior work. The code of our work is available at https://github.com/columbia/MetricLearningAdversarial_Robustness.

Although deep learning models perform remarkably well across a range of tasks such as language translation and object recognition, it remains unclear what high-level logic, if any, they follow. Understanding this logic may lead to more transparency, better model design, and faster experimentation. Recent machine learning research has leveraged statistical methods to identify hidden units that behave (e.g., activate) similarly to human understandable logic, but those analyses require considerable manual effort. Our insight is that many of those studies follow a common analysis pattern, and therefore there is opportunity to provide a declarative abstraction to easily express, execute and optimize them. This paper describes DeepBase, a system to inspect neural network behaviors through a unified interface. We model logic with user-provided hypothesis functions that annotate the data with high-level labels (e.g., part-of-speech tags, image captions). DeepBase lets users quickly identify individual or groups of units that have strong statistical dependencies with desired hypotheses. We discuss how DeepBase can express existing analyses, propose a set of simple and effective optimizations to speed up a standard Python implementation by up to 72x, and reproduce recent studies from the NLP literature.

This paper introduces a video dataset of spatio-temporally localized Atomic Visual Actions (AVA). The AVA dataset densely annotates 80 atomic visual actions in 437 15-minute video clips, where actions are localized in space and time, resulting in 1.59M action labels with multiple labels per person occurring frequently. The key characteristics of our dataset are: (1) the definition of atomic visual actions, rather than composite actions; (2) precise spatio-temporal annotations with possibly multiple annotations for each person; (3) exhaustive annotation of these atomic actions over 15-minute video clips; (4) people temporally linked across consecutive segments; and (5) using movies to gather a varied set of action representations. This departs from existing datasets for spatio-temporal action recognition, which typically provide sparse annotations for composite actions in short video clips. AVA, with its realistic scene and action complexity, exposes the intrinsic difficulty of action recognition. To benchmark this, we present a novel approach for action localization that builds upon the current state-of-the-art methods, and demonstrates better performance on JHMDB and UCF101-24 categories. While setting a new state of the art on existing datasets, the overall results on AVA are low at 15.8% mAP, underscoring the need for developing new approaches for video understanding.

People can recognize scenes across many different modalities beyond natural images. In this paper, we investigate how to learn cross-modal scene representations that transfer across modalities. To study this problem, we introduce a new cross-modal scene dataset. While convolutional neural networks can categorize scenes well, they also learn an intermediate representation not aligned across modalities, which is undesirable for cross-modal transfer applications. We present methods to regularize cross-modal convolutional neural networks so that they have a shared representation that is agnostic of the modality. Our experiments suggest that our scene representation can help transfer representations across modalities for retrieval. Moreover, our visualizations suggest that units emerge in the shared representation that tend to activate on consistent concepts independently of the modality.

We learn models to generate the immediate future in video. This problem has two main challenges. Firstly, since the future is uncertain, models should be multi-modal, which can be difficult to learn. Secondly, since the future is similar to the past, models store low-level details, which complicates learning of high-level semantics. We propose a framework to tackle both of these challenges. We present a model that generates the future by transforming pixels in the past. Our approach explicitly disentangles the models memory from the prediction, which helps the model learn desirable invariances. Experiments suggest that this model can generate short videos of plausible futures. We believe predictive models have many applications in robotics, health-care, and video understanding.

Following the gaze of people inside videos is an important signal for understanding people and their actions. In this paper, we present an approach for following gaze in video by predicting where a person (in the video) is looking even when the object is in a different frame. We collect VideoGaze, a new dataset which we use as a benchmark to both train and evaluate models. Given one frame with a person in it, our model estimates a density for gaze location in every frame and the probability that the person is looking in that particular frame. A key aspect of our approach is an end-to-end model that jointly estimates: saliency, gaze pose, and geometric relationships between views while only using gaze as supervision. Visualizations suggest that the model learns to internally solve these intermediate tasks automatically without additional supervision. Experiments show that our approach follows gaze in video better than existing approaches, enabling a richer understanding of human activities in video.

People can recognize scenes across many different modalities beyond natural images. In this paper, we investigate how to learn cross-modal scene representations that transfer across modalities. To study this problem, we introduce a new cross-modal scene dataset. While convolutional neural networks can categorize cross-modal scenes well, they also learn an intermediate representation not aligned across modalities, which is undesirable for crossmodal transfer applications. We present methods to regularize cross-modal convolutional neural networks so that they have a shared representation that is agnostic of the modality. Our experiments suggest that our scene representation can help transfer representations across modalities for retrieval. Moreover, our visualizations suggest that units emerge in the shared representation that tend to activate on consistent concepts independently of the modality.

Understanding human actions is a key problem in computer vision. However, recognizing actions is only the first step of understanding what a person is doing. In this paper, we introduce the problem of predicting why a person has performed an action in images. This problem has many applications in human activity understanding, such as anticipating or explaining an action. To study this problem, we introduce a new dataset of people performing actions annotated with likely motivations. However, the information in an image alone may not be sufficient to automatically solve this task. Since humans can rely on their lifetime of experiences to infer motivation, we propose to give computer vision systems access to some of these experiences by using recently developed natural language models to mine knowledge stored in massive amounts of text. While we are still far away from fully understanding motivation, our results suggest that transferring knowledge from language into vision can help machines understand why people in images might be performing an action.

Anticipating actions and objects before they start or appear is a difficult problem in computer vision with several real-world applications. This task is challenging partly because it requires leveraging extensive knowledge of the world that is difficult to write down. We believe that a promising resource for efficiently learning this knowledge is through readily available unlabeled video. We present a framework that capitalizes on temporal structure in unlabeled video to learn to anticipate human actions and objects. The key idea behind our approach is that we can train deep networks to predict the visual representation of images in the future. Visual representations are a promising prediction target because they encode images at a higher semantic level than pixels yet are automatic to compute. We then apply recognition algorithms on our predicted representation to anticipate objects and actions. We experimentally validate this idea on two datasets, anticipating actions one second in the future and objects five seconds in the future.

We learn rich natural sound representations by capitalizing on large amounts of unlabeled sound data collected in the wild. We leverage the natural synchronization between vision and sound to learn an acoustic representation using two-million unlabeled videos. Unlabeled video has the advantage that it can be economically acquired at massive scales, yet contains useful signals about natural sound. We propose a student-teacher training procedure which transfers discriminative visual knowledge from well established visual recognition models into the sound modality using unlabeled video as a bridge. Our sound representation yields significant performance improvements over the state-of-the-art results on standard benchmarks for acoustic scene/object classification. Visualizations suggest some high-level semantics automatically emerge in the sound network, even though it is trained without ground truth labels.

We capitalize on large amounts of unlabeled video in order to learn a model of scene dynamics for both video recognition tasks (e.g. action classification) and video generation tasks (e.g. future prediction). We propose a generative adversarial network for video with a spatio-temporal convolutional architecture that untangles the scene's foreground from the background. Experiments suggest this model can generate tiny videos up to a second at full frame rate better than simple baselines, and we show its utility at predicting plausible futures of static images. Moreover, experiments and visualizations show the model internally learns useful features for recognizing actions with minimal supervision, suggesting scene dynamics are a promising signal for representation learning. We believe generative video models can impact many applications in video understanding and simulation.

Humans have the remarkable ability to follow the gaze of other people to identify what they are looking at. Following eye gaze, or gaze-following, is an important ability that allows us to understand what other people are thinking, the actions they are performing, and even predict what they might do next. Despite the importance of this topic, this problem has only been studied in limited scenarios within the computer vision community. In this paper, we propose a deep neural network-based approach for gaze-following and a new benchmark dataset for thorough evaluation. Given an image and the location of a head, our approach follows the gaze of the person and identifies the object being looked at. After training, the network is able to discover how to extract head pose and gaze orientation, and to select objects in the scene that are in the predicted line of sight and likely to be looked at (such as televisions, balls and food). The quantitative evaluation shows that our approach produces reliable results, even when viewing only the back of the head. While our method outperforms several baseline approaches, we are still far from reaching human performance at this task. Overall, we believe that this is a challenging and important task that deserves more attention from the community.

Although the human visual system can recognize many concepts under challengingconditions, it still has some biases. In this paper, we investigate whether wecan extract these biases and transfer them into a machine recognition system.We introduce a novel method that, inspired by well-known tools in humanpsychophysics, estimates the biases that the human visual system might use forrecognition, but in computer vision feature spaces. Our experiments aresurprising, and suggest that classifiers from the human visual system can betransferred into a machine with some success. Since these classifiers seem tocapture favorable biases in the human visual system, we further present an SVMformulation that constrains the orientation of the SVM hyperplane to agree withthe bias from human visual system. Our results suggest that transferring thishuman bias into machines may help object recognition systems generalize acrossdatasets and perform better when very little training data is available.

We introduce algorithms to visualize feature spaces used by object detectors. The tools in this paper allow a human to put on 'HOG goggles' and perceive the visual world as a HOG based object detector sees it. We found that these visualizations allow us to analyze object detection systems in new ways and gain new insight into the detector's failures. For example, when we visualize the features for high scoring false alarms, we discovered that, although they are clearly wrong in image space, they do look deceptively similar to true positives in feature space. This result suggests that many of these false alarms are caused by our choice of feature space, and indicates that creating a better learning algorithm or building bigger datasets is unlikely to correct these errors. By visualizing feature spaces, we can gain a more intuitive understanding of our detection systems.

(Work performed while at UC Irvine) Datasets for training object recognition systems are steadily growing in size. This paper investigates the question of whether existing detectors will continue to improve as data grows, or if models are close to saturating due to limited model complexity and the Bayes risk associated with the feature spaces in which they operate. We focus on the popular paradigm of scanning-window templates defined on oriented gradient fea-tures, trained with discriminative classifiers. We investigate the performance of mixtures of templates as a function of the number of templates (complexity) and the amount of training data. We find that additional data does help, but only with correct regularization and treatment of noisy examples or “outliers ” in the training data. Surprisingly, the per-formance of problem domain-agnostic mixture models appears to saturate quickly (∼10 templates and ∼100 positive training examples per template). However, compositional mixtures (implemented via composed parts) give much better performance because they share parameters among templates, and can synthesize new templates not encountered during training. This suggests there is still room to improve performance with linear classifiers and the existing feature space by improved representations and learning algo-rithms. 1

Summary form only given. We present a concept for automatic construction site monitoring by taking into account 4D information (3D over time), that is acquired from highly-overlapping digital aerial images. On the one hand today's maturity of flying micro aerial vehicles (MAVs) enables a low-cost and an efficient image acquisition of high-quality data that maps construction sites entirely from many varying viewpoints. On the other hand, due to low-noise sensors and high redundancy in the image data, recent developments in 3D reconstruction workflows have benefited the automatic computation of accurate and dense 3D scene information. Having both an inexpensive high-quality image acquisition and an efficient 3D analysis workflow enables monitoring, documentation and visualization of observed sites over time with short intervals. Relating acquired 4D site observations, composed of color, texture, geometry over time, largely supports automated methods toward full scene understanding, the acquisition of both the change and the construction site's progress.

We introduce a new large-scale video dataset designed to assess the performance of diverse visual event recognition algorithms with a focus on continuous visual event recognition (CVER) in outdoor areas with wide coverage. Previous datasets for action recognition are unrealistic for real-world surveillance because they consist of short clips showing one action by one individual [15, 8]. Datasets have been developed for movies [11] and sports [12], but, these actions and scene conditions do not apply effectively to surveillance videos. Our dataset consists of many outdoor scenes with actions occurring naturally by non-actors in continuously captured videos of the real world. The dataset includes large numbers of instances for 23 event types distributed throughout 29 hours of video. This data is accompanied by detailed annotations which include both moving object tracks and event examples, which will provide solid basis for large-scale evaluation. Additionally, we propose different types of evaluation modes for visual recognition tasks and evaluation metrics along with our preliminary experimental results. We believe that this dataset will stimulate diverse aspects of computer vision research and help us to advance the CVER tasks in the years ahead.

We introduce a novel active learning framework for video annotation. By judiciously choosing which frames a user should annotate, we can obtain highly accurate tracks with minimal user effort. We cast this problem as one of active learning, and show that we can obtain excellent performance by querying frames that, if annotated, would produce a large expected change in the estimated object track. We implement a constrained tracker and compute the expected change for putative annotations with efficient dynamic programming algorithms. We demonstrate our framework on four datasets, including two benchmark datasets constructed with key frame annotations obtained by Amazon Mechanical Turk. Our results indicate that we could obtain equivalent labels for a small fraction of the original cost.