João F. Henriques

Stephanie Milani, Anssi Kanervisto, Karolis Ramanauskas et al. · berkeley

To facilitate research in the direction of fine-tuning foundation models from human feedback, we held the MineRL BASALT Competition on Fine-Tuning from Human Feedback at NeurIPS 2022. The BASALT challenge asks teams to compete to develop algorithms to solve tasks with hard-to-specify reward functions in Minecraft. Through this competition, we aimed to promote the development of algorithms that use human feedback as channels to learn the desired behavior. We describe the competition and provide an overview of the top solutions. We conclude by discussing the impact of the competition and future directions for improvement.

Yash Bhalgat, Iro Laina, João F. Henriques et al.

Instance segmentation in 3D is a challenging task due to the lack of large-scale annotated datasets. In this paper, we show that this task can be addressed effectively by leveraging instead 2D pre-trained models for instance segmentation. We propose a novel approach to lift 2D segments to 3D and fuse them by means of a neural field representation, which encourages multi-view consistency across frames. The core of our approach is a slow-fast clustering objective function, which is scalable and well-suited for scenes with a large number of objects. Unlike previous approaches, our method does not require an upper bound on the number of objects or object tracking across frames. To demonstrate the scalability of the slow-fast clustering, we create a new semi-realistic dataset called the Messy Rooms dataset, which features scenes with up to 500 objects per scene. Our approach outperforms the state-of-the-art on challenging scenes from the ScanNet, Hypersim, and Replica datasets, as well as on our newly created Messy Rooms dataset, demonstrating the effectiveness and scalability of our slow-fast clustering method.

Yan Xia, Mariia Gladkova, Rui Wang et al.

Place recognition based on point clouds (LiDAR) is an important component for autonomous robots or self-driving vehicles. Current SOTA performance is achieved on accumulated LiDAR submaps using either point-based or voxel-based structures. While voxel-based approaches nicely integrate spatial context across multiple scales, they do not exhibit the local precision of point-based methods. As a result, existing methods struggle with fine-grained matching of subtle geometric features in sparse single-shot Li- DAR scans. To overcome these limitations, we propose CASSPR as a method to fuse point-based and voxel-based approaches using cross attention transformers. CASSPR leverages a sparse voxel branch for extracting and aggregating information at lower resolution and a point-wise branch for obtaining fine-grained local information. CASSPR uses queries from one branch to try to match structures in the other branch, ensuring that both extract self-contained descriptors of the point cloud (rather than one branch dominating), but using both to inform the output global descriptor of the point cloud. Extensive experiments show that CASSPR surpasses the state-of-the-art by a large margin on several datasets (Oxford RobotCar, TUM, USyd). For instance, it achieves AR@1 of 85.6% on the TUM dataset, surpassing the strongest prior model by ~15%. Our code is publicly available.

Tim Franzmeyer, Philip H. S. Torr, João F. Henriques

We study how an autonomous agent learns to perform a task from demonstrations in a different domain, such as a different environment or different agent. Such cross-domain imitation learning is required to, for example, train an artificial agent from demonstrations of a human expert. We propose a scalable framework that enables cross-domain imitation learning without access to additional demonstrations or further domain knowledge. We jointly train the learner agent's policy and learn a mapping between the learner and expert domains with adversarial training. We effect this by using a mutual information criterion to find an embedding of the expert's state space that contains task-relevant information and is invariant to domain specifics. This step significantly simplifies estimating the mapping between the learner and expert domains and hence facilitates end-to-end learning. We demonstrate successful transfer of policies between considerably different domains, without extra supervision such as additional demonstrations, and in situations where other methods fail.

Eldar Insafutdinov, Dylan Campbell, João F. Henriques et al.

We present a method for the accurate 3D reconstruction of partly-symmetric objects. We build on the strengths of recent advances in neural reconstruction and rendering such as Neural Radiance Fields (NeRF). A major shortcoming of such approaches is that they fail to reconstruct any part of the object which is not clearly visible in the training image, which is often the case for in-the-wild images and videos. When evidence is lacking, structural priors such as symmetry can be used to complete the missing information. However, exploiting such priors in neural rendering is highly non-trivial: while geometry and non-reflective materials may be symmetric, shadows and reflections from the ambient scene are not symmetric in general. To address this, we apply a soft symmetry constraint to the 3D geometry and material properties, having factored appearance into lighting, albedo colour and reflectivity. We evaluate our method on the recently introduced CO3D dataset, focusing on the car category due to the challenge of reconstructing highly-reflective materials. We show that it can reconstruct unobserved regions with high fidelity and render high-quality novel view images.

Yash Bhalgat, Vadim Tschernezki, Iro Laina et al.

Egocentric videos present unique challenges for 3D scene understanding due to rapid camera motion, frequent object occlusions, and limited object visibility. This paper introduces a novel approach to instance segmentation and tracking in first-person video that leverages 3D awareness to overcome these obstacles. Our method integrates scene geometry, 3D object centroid tracking, and instance segmentation to create a robust framework for analyzing dynamic egocentric scenes. By incorporating spatial and temporal cues, we achieve superior performance compared to state-of-the-art 2D approaches. Extensive evaluations on the challenging EPIC Fields dataset demonstrate significant improvements across a range of tracking and segmentation consistency metrics. Specifically, our method outperforms the next best performing approach by $7$ points in Association Accuracy (AssA) and $4.5$ points in IDF1 score, while reducing the number of ID switches by $73\%$ to $80\%$ across various object categories. Leveraging our tracked instance segmentations, we showcase downstream applications in 3D object reconstruction and amodal video object segmentation in these egocentric settings.

Nazir Nayal, Mısra Yavuz, João F. Henriques et al.

Standard semantic segmentation models owe their success to curated datasets with a fixed set of semantic categories, without contemplating the possibility of identifying unknown objects from novel categories. Existing methods in outlier detection suffer from a lack of smoothness and objectness in their predictions, due to limitations of the per-pixel classification paradigm. Furthermore, additional training for detecting outliers harms the performance of known classes. In this paper, we explore another paradigm with region-level classification to better segment unknown objects. We show that the object queries in mask classification tend to behave like one \vs all classifiers. Based on this finding, we propose a novel outlier scoring function called RbA by defining the event of being an outlier as being rejected by all known classes. Our extensive experiments show that mask classification improves the performance of the existing outlier detection methods, and the best results are achieved with the proposed RbA. We also propose an objective to optimize RbA using minimal outlier supervision. Further fine-tuning with outliers improves the unknown performance, and unlike previous methods, it does not degrade the inlier performance.

Samuel Albanie, Liliane Momeni, João F. Henriques · cambridge

Driven by recent advances AI, we passengers are entering a golden age of scientific discovery. But golden for whom? Confronting our insecurity that others may beat us to the most acclaimed breakthroughs of the era, we propose a novel solution to the long-standing personal credit assignment problem to ensure that it is golden for us. At the heart of our approach is a pip-to-the-post algorithm that assures adulatory Wikipedia pages without incurring the substantial capital and career risks of pursuing high impact science with conventional research methodologies. By leveraging the meta trend of leveraging large language models for everything, we demonstrate the unparalleled potential of our algorithm to scoop groundbreaking findings with the insouciance of a seasoned researcher at a dessert buffet.

Felipe Nuti, Tim Franzmeyer, João F. Henriques

Diffusion models have achieved remarkable results in image generation, and have similarly been used to learn high-performing policies in sequential decision-making tasks. Decision-making diffusion models can be trained on lower-quality data, and then be steered with a reward function to generate near-optimal trajectories. We consider the problem of extracting a reward function by comparing a decision-making diffusion model that models low-reward behavior and one that models high-reward behavior; a setting related to inverse reinforcement learning. We first define the notion of a relative reward function of two diffusion models and show conditions under which it exists and is unique. We then devise a practical learning algorithm for extracting it by aligning the gradients of a reward function -- parametrized by a neural network -- to the difference in outputs of both diffusion models. Our method finds correct reward functions in navigation environments, and we demonstrate that steering the base model with the learned reward functions results in significantly increased performance in standard locomotion benchmarks. Finally, we demonstrate that our approach generalizes beyond sequential decision-making by learning a reward-like function from two large-scale image generation diffusion models. The extracted reward function successfully assigns lower rewards to harmful images.

Yan Xia, Letian Shi, Zifeng Ding et al.

We tackle the problem of 3D point cloud localization based on a few natural linguistic descriptions and introduce a novel neural network, Text2Loc, that fully interprets the semantic relationship between points and text. Text2Loc follows a coarse-to-fine localization pipeline: text-submap global place recognition, followed by fine localization. In global place recognition, relational dynamics among each textual hint are captured in a hierarchical transformer with max-pooling (HTM), whereas a balance between positive and negative pairs is maintained using text-submap contrastive learning. Moreover, we propose a novel matching-free fine localization method to further refine the location predictions, which completely removes the need for complicated text-instance matching and is lighter, faster, and more accurate than previous methods. Extensive experiments show that Text2Loc improves the localization accuracy by up to $2\times$ over the state-of-the-art on the KITTI360Pose dataset. Our project page is publicly available at \url{https://yan-xia.github.io/projects/text2loc/}.

Tim Franzmeyer, Stephen McAleer, João F. Henriques et al.

Autonomous agents deployed in the real world need to be robust against adversarial attacks on sensory inputs. Robustifying agent policies requires anticipating the strongest attacks possible. We demonstrate that existing observation-space attacks on reinforcement learning agents have a common weakness: while effective, their lack of information-theoretic detectability constraints makes them detectable using automated means or human inspection. Detectability is undesirable to adversaries as it may trigger security escalations. We introduce ε-illusory, a novel form of adversarial attack on sequential decision-makers that is both effective and of ε-bounded statistical detectability. We propose a novel dual ascent algorithm to learn such attacks end-to-end. Compared to existing attacks, we empirically find ε-illusory to be significantly harder to detect with automated methods, and a small study with human participants (IRB approval under reference R84123/RE001) suggests they are similarly harder to detect for humans. Our findings suggest the need for better anomaly detectors, as well as effective hardware- and system-level defenses. The project website can be found at https://tinyurl.com/illusory-attacks.

Dominik A. Kloepfer, Dylan Campbell, João F. Henriques

An important challenge for autonomous agents such as robots is to maintain a spatially and temporally consistent model of the world. It must be maintained through occlusions, previously-unseen views, and long time horizons (e.g., loop closure and re-identification). It is still an open question how to train such a versatile neural representation without supervision. We start from the idea that the training objective can be framed as a patch retrieval problem: given an image patch in one view of a scene, we would like to retrieve (with high precision and recall) all patches in other views that map to the same real-world location. One drawback is that this objective does not promote reusability of features: by being unique to a scene (achieving perfect precision/recall), a representation will not be useful in the context of other scenes. We find that it is possible to balance retrieval and reusability by constructing the retrieval set carefully, leaving out patches that map to far-away locations. Similarly, we can easily regulate the scale of the learned features (e.g., points, objects, or rooms) by adjusting the spatial tolerance for considering a retrieval to be positive. We optimize for (smooth) Average Precision (AP), in a single unified ranking-based objective. This objective also doubles as a criterion for choosing landmarks or keypoints, as patches with high AP. We show results creating sparse, multi-scale, semantic spatial maps composed of highly identifiable landmarks, with applications in landmark retrieval, localization, semantic segmentation and instance segmentation.

Andreea-Maria Oncescu, João F. Henriques, A. Sophia Koepke

Recent advancements in machine learning have fueled research on multimodal tasks, such as for instance text-to-video and text-to-audio retrieval. These tasks require models to understand the semantic content of video and audio data, including objects, and characters. The models also need to learn spatial arrangements and temporal relationships. In this work, we analyse the temporal ordering of sounds, which is an understudied problem in the context of text-to-audio retrieval. In particular, we dissect the temporal understanding capabilities of a state-of-the-art model for text-to-audio retrieval on the AudioCaps and Clotho datasets. Additionally, we introduce a synthetic text-audio dataset that provides a controlled setting for evaluating temporal capabilities of recent models. Lastly, we present a loss function that encourages text-audio models to focus on the temporal ordering of events. Code and data are available at https://www.robots.ox.ac.uk/~vgg/research/audio-retrieval/dtu/.

Shu Ishida, João F. Henriques

This work compares ways of extending Reinforcement Learning algorithms to Partially Observed Markov Decision Processes (POMDPs) with options. One view of options is as temporally extended action, which can be realized as a memory that allows the agent to retain historical information beyond the policy's context window. While option assignment could be handled using heuristics and hand-crafted objectives, learning temporally consistent options and associated sub-policies without explicit supervision is a challenge. Two algorithms, PPOEM and SOAP, are proposed and studied in depth to address this problem. PPOEM applies the forward-backward algorithm (for Hidden Markov Models) to optimize the expected returns for an option-augmented policy. However, this learning approach is unstable during on-policy rollouts. It is also unsuited for learning causal policies without the knowledge of future trajectories, since option assignments are optimized for offline sequences where the entire episode is available. As an alternative approach, SOAP evaluates the policy gradient for an optimal option assignment. It extends the concept of the generalized advantage estimation (GAE) to propagate option advantages through time, which is an analytical equivalent to performing temporal back-propagation of option policy gradients. This option policy is only conditional on the history of the agent, not future actions. Evaluated against competing baselines, SOAP exhibited the most robust performance, correctly discovering options for POMDP corridor environments, as well as on standard benchmarks including Atari and MuJoCo, outperforming PPOEM, as well as LSTM and Option-Critic baselines. The open-sourced code is available at https://github.com/shuishida/SoapRL.

Shu Ishida, Gianluca Corrado, George Fedoseev et al.

We propose LangProp, a framework for iteratively optimizing code generated by large language models (LLMs), in both supervised and reinforcement learning settings. While LLMs can generate sensible coding solutions zero-shot, they are often sub-optimal. Especially for code generation tasks, it is likely that the initial code will fail on certain edge cases. LangProp automatically evaluates the code performance on a dataset of input-output pairs, catches any exceptions, and feeds the results back to the LLM in the training loop, so that the LLM can iteratively improve the code it generates. By adopting a metric- and data-driven training paradigm for this code optimization procedure, one could easily adapt findings from traditional machine learning techniques such as imitation learning, DAgger, and reinforcement learning. We show LangProp's applicability to general domains such as Sudoku and CartPole, as well as demonstrate the first proof of concept of automated code optimization for autonomous driving in CARLA. We show that LangProp can generate interpretable and transparent policies that can be verified and improved in a metric- and data-driven way. Our code is available at https://github.com/shuishida/LangProp.

A. Sophia Koepke, Andreea-Maria Oncescu, João F. Henriques et al.

The objectives of this work are cross-modal text-audio and audio-text retrieval, in which the goal is to retrieve the audio content from a pool of candidates that best matches a given written description and vice versa. Text-audio retrieval enables users to search large databases through an intuitive interface: they simply issue free-form natural language descriptions of the sound they would like to hear. To study the tasks of text-audio and audio-text retrieval, which have received limited attention in the existing literature, we introduce three challenging new benchmarks. We first construct text-audio and audio-text retrieval benchmarks from the AudioCaps and Clotho audio captioning datasets. Additionally, we introduce the SoundDescs benchmark, which consists of paired audio and natural language descriptions for a diverse collection of sounds that are complementary to those found in AudioCaps and Clotho. We employ these three benchmarks to establish baselines for cross-modal text-audio and audio-text retrieval, where we demonstrate the benefits of pre-training on diverse audio tasks. We hope that our benchmarks will inspire further research into audio retrieval with free-form text queries. Code, audio features for all datasets used, and the SoundDescs dataset are publicly available at https://github.com/akoepke/audio-retrieval-benchmark.

Mandela Patrick, Dylan Campbell, Yuki M. Asano et al.

In video transformers, the time dimension is often treated in the same way as the two spatial dimensions. However, in a scene where objects or the camera may move, a physical point imaged at one location in frame $t$ may be entirely unrelated to what is found at that location in frame $t+k$. These temporal correspondences should be modeled to facilitate learning about dynamic scenes. To this end, we propose a new drop-in block for video transformers -- trajectory attention -- that aggregates information along implicitly determined motion paths. We additionally propose a new method to address the quadratic dependence of computation and memory on the input size, which is particularly important for high resolution or long videos. While these ideas are useful in a range of settings, we apply them to the specific task of video action recognition with a transformer model and obtain state-of-the-art results on the Kinetics, Something--Something V2, and Epic-Kitchens datasets. Code and models are available at: https://github.com/facebookresearch/Motionformer

Xu Ji, João F. Henriques, Andrea Vedaldi

We present a novel clustering objective that learns a neural network classifier from scratch, given only unlabelled data samples. The model discovers clusters that accurately match semantic classes, achieving state-of-the-art results in eight unsupervised clustering benchmarks spanning image classification and segmentation. These include STL10, an unsupervised variant of ImageNet, and CIFAR10, where we significantly beat the accuracy of our closest competitors by 6.6 and 9.5 absolute percentage points respectively. The method is not specialised to computer vision and operates on any paired dataset samples; in our experiments we use random transforms to obtain a pair from each image. The trained network directly outputs semantic labels, rather than high dimensional representations that need external processing to be usable for semantic clustering. The objective is simply to maximise mutual information between the class assignments of each pair. It is easy to implement and rigorously grounded in information theory, meaning we effortlessly avoid degenerate solutions that other clustering methods are susceptible to. In addition to the fully unsupervised mode, we also test two semi-supervised settings. The first achieves 88.8% accuracy on STL10 classification, setting a new global state-of-the-art over all existing methods (whether supervised, semi-supervised or unsupervised). The second shows robustness to 90% reductions in label coverage, of relevance to applications that wish to make use of small amounts of labels. github.com/xu-ji/IIC

João F. Henriques, Sebastien Ehrhardt, Samuel Albanie et al.

We propose a fast second-order method that can be used as a drop-in replacement for current deep learning solvers. Compared to stochastic gradient descent (SGD), it only requires two additional forward-mode automatic differentiation operations per iteration, which has a computational cost comparable to two standard forward passes and is easy to implement. Our method addresses long-standing issues with current second-order solvers, which invert an approximate Hessian matrix every iteration exactly or by conjugate-gradient methods, a procedure that is both costly and sensitive to noise. Instead, we propose to keep a single estimate of the gradient projected by the inverse Hessian matrix, and update it once per iteration. This estimate has the same size and is similar to the momentum variable that is commonly used in SGD. No estimate of the Hessian is maintained. We first validate our method, called CurveBall, on small problems with known closed-form solutions (noisy Rosenbrock function and degenerate 2-layer linear networks), where current deep learning solvers seem to struggle. We then train several large models on CIFAR and ImageNet, including ResNet and VGG-f networks, where we demonstrate faster convergence with no hyperparameter tuning. Code is available.

João F. Henriques, Rui Caseiro, Pedro Martins et al.

The core component of most modern trackers is a discriminative classifier, tasked with distinguishing between the target and the surrounding environment. To cope with natural image changes, this classifier is typically trained with translated and scaled sample patches. Such sets of samples are riddled with redundancies -- any overlapping pixels are constrained to be the same. Based on this simple observation, we propose an analytic model for datasets of thousands of translated patches. By showing that the resulting data matrix is circulant, we can diagonalize it with the Discrete Fourier Transform, reducing both storage and computation by several orders of magnitude. Interestingly, for linear regression our formulation is equivalent to a correlation filter, used by some of the fastest competitive trackers. For kernel regression, however, we derive a new Kernelized Correlation Filter (KCF), that unlike other kernel algorithms has the exact same complexity as its linear counterpart. Building on it, we also propose a fast multi-channel extension of linear correlation filters, via a linear kernel, which we call Dual Correlation Filter (DCF). Both KCF and DCF outperform top-ranking trackers such as Struck or TLD on a 50 videos benchmark, despite running at hundreds of frames-per-second, and being implemented in a few lines of code (Algorithm 1). To encourage further developments, our tracking framework was made open-source.

Yash Bhalgat, Iro Laina, João F. Henriques et al.

Understanding complex scenes at multiple levels of abstraction remains a formidable challenge in computer vision. To address this, we introduce Nested Neural Feature Fields (N2F2), a novel approach that employs hierarchical supervision to learn a single feature field, wherein different dimensions within the same high-dimensional feature encode scene properties at varying granularities. Our method allows for a flexible definition of hierarchies, tailored to either the physical dimensions or semantics or both, thereby enabling a comprehensive and nuanced understanding of scenes. We leverage a 2D class-agnostic segmentation model to provide semantically meaningful pixel groupings at arbitrary scales in the image space, and query the CLIP vision-encoder to obtain language-aligned embeddings for each of these segments. Our proposed hierarchical supervision method then assigns different nested dimensions of the feature field to distill the CLIP embeddings using deferred volumetric rendering at varying physical scales, creating a coarse-to-fine representation. Extensive experiments show that our approach outperforms the state-of-the-art feature field distillation methods on tasks such as open-vocabulary 3D segmentation and localization, demonstrating the effectiveness of the learned nested feature field.

Yichao Liang, Nishanth Kumar, Hao Tang et al.

Broadly intelligent agents should form task-specific abstractions that selectively expose the essential elements of a task, while abstracting away the complexity of the raw sensorimotor space. In this work, we present Neuro-Symbolic Predicates, a first-order abstraction language that combines the strengths of symbolic and neural knowledge representations. We outline an online algorithm for inventing such predicates and learning abstract world models. We compare our approach to hierarchical reinforcement learning, vision-language model planning, and symbolic predicate invention approaches, on both in- and out-of-distribution tasks across five simulated robotic domains. Results show that our approach offers better sample complexity, stronger out-of-distribution generalization, and improved interpretability.

Mark C. Eid, Pak-Hei Yeung, Madeleine K. Wyburd et al.

Two-dimensional (2D) freehand ultrasonography is one of the most commonly used medical imaging modalities, particularly in obstetrics and gynaecology. However, it only captures 2D cross-sectional views of inherently 3D anatomies, losing valuable contextual information. As an alternative to requiring costly and complex 3D ultrasound scanners, 3D volumes can be constructed from 2D scans using machine learning. However this usually requires long computational time. Here, we propose RapidVol: a neural representation framework to speed up slice-to-volume ultrasound reconstruction. We use tensor-rank decomposition, to decompose the typical 3D volume into sets of tri-planes, and store those instead, as well as a small neural network. A set of 2D ultrasound scans, with their ground truth (or estimated) 3D position and orientation (pose) is all that is required to form a complete 3D reconstruction. Reconstructions are formed from real fetal brain scans, and then evaluated by requesting novel cross-sectional views. When compared to prior approaches based on fully implicit representation (e.g. neural radiance fields), our method is over 3x quicker, 46% more accurate, and if given inaccurate poses is more robust. Further speed-up is also possible by reconstructing from a structural prior rather than from scratch.

Yan Xia, Zhendong Li, Yun-Jin Li et al.

To date, most place recognition methods focus on single-modality retrieval. While they perform well in specific environments, cross-modal methods offer greater flexibility by allowing seamless switching between map and query sources. It also promises to reduce computation requirements by having a unified model, and achieving greater sample efficiency by sharing parameters. In this work, we develop a universal solution to place recognition, UniLoc, that works with any single query modality (natural language, image, or point cloud). UniLoc leverages recent advances in large-scale contrastive learning, and learns by matching hierarchically at two levels: instance-level matching and scene-level matching. Specifically, we propose a novel Self-Attention based Pooling (SAP) module to evaluate the importance of instance descriptors when aggregated into a place-level descriptor. Experiments on the KITTI-360 dataset demonstrate the benefits of cross-modality for place recognition, achieving superior performance in cross-modal settings and competitive results also for uni-modal scenarios. Our project page is publicly available at https://yan-xia.github.io/projects/UniLoc/.

Mark C. Eid, Ana I. L. Namburete, João F. Henriques

Ultrasound imaging is widely used due to its safety, affordability, and real-time capabilities, but its 2D interpretation is highly operator-dependent, leading to variability and increased cognitive demand. 2D-to-3D reconstruction mitigates these challenges by providing standardized volumetric views, yet existing methods are often computationally expensive, memory-intensive, or incompatible with ultrasound physics. We introduce UltraGauss: the first ultrasound-specific Gaussian Splatting framework, extending view synthesis techniques to ultrasound wave propagation. Unlike conventional perspective-based splatting, UltraGauss models probe-plane intersections in 3D, aligning with acoustic image formation. We derive an efficient rasterization boundary formulation for GPU parallelization and introduce a numerically stable covariance parametrization, improving computational efficiency and reconstruction accuracy. On real clinical ultrasound data, UltraGauss achieves state-of-the-art reconstructions in 5 minutes, and reaching 0.99 SSIM within 20 minutes on a single GPU. A survey of expert clinicians confirms UltraGauss' reconstructions are the most realistic among competing methods. Our CUDA implementation will be released upon publication.

Yan Xia, Yunxiang Lu, Rui Song et al.

We tackle the problem of localizing traffic cameras within a 3D reference map and propose a novel image-to-point cloud registration (I2P) method, TrafficLoc, in a coarse-tofine matching fashion. To overcome the lack of large-scale real-world intersection datasets, we first introduce Carla Intersection, a new simulated dataset with 75 urban and rural intersections in Carla. We find that current I2P methods struggle with cross-modal matching under large viewpoint differences, especially at traffic intersections. TrafficLoc thus employs a novel Geometry-guided Attention Loss (GAL) to focus only on the corresponding geometric regions under different viewpoints during 2D-3D feature fusion. To address feature inconsistency in paired image patch-point groups, we further propose Inter-intra Contrastive Learning (ICL) to enhance separating 2D patch/3D group features within each intra-modality and introduce Dense Training Alignment (DTA) with soft-argmax for improving position regression. Extensive experiments show our TrafficLoc greatly improves the performance over the SOTA I2P methods (up to 86%) on Carla Intersection and generalizes well to real-world data. TrafficLoc also achieves new SOTA performance on KITTI and NuScenes datasets, demonstrating the superiority across both in-vehicle and traffic cameras. Our project page is publicly available at https://tum-luk.github.io/projects/trafficloc/.

Marian Longa, João F. Henriques

Learning interpretable representations of visual data is an important challenge, to make machines' decisions understandable to humans and to improve generalisation outside of the training distribution. To this end, we propose a deep learning framework where one can specify nonlinear priors for videos (e.g. of Newtonian physics) that allow the model to learn interpretable latent variables and use these to generate videos of hypothetical scenarios not observed at training time. We do this by extending the Variational Auto-Encoder (VAE) prior from a simple isotropic Gaussian to an arbitrary nonlinear temporal Additive Noise Model (ANM), which can describe a large number of processes (e.g. Newtonian physics). We propose a novel linearization method that constructs a Gaussian Mixture Model (GMM) approximating the prior, and derive a numerically stable Monte Carlo estimate of the KL divergence between the posterior and prior GMMs. We validate the method on different real-world physics videos including a pendulum, a mass on a spring, a falling object and a pulsar (rotating neutron star). We specify a physical prior for each experiment and show that the correct variables are learned. Once a model is trained, we intervene on it to change different physical variables (such as oscillation amplitude or adding air drag) to generate physically correct videos of hypothetical scenarios that were not observed previously.

Marian Longa, João F. Henriques

Unsupervised object detection using deep neural networks is typically a difficult problem with few to no guarantees about the learned representation. In this work we present the first unsupervised object detection method that is theoretically guaranteed to recover the true object positions up to quantifiable small shifts. We develop an unsupervised object detection architecture and prove that the learned variables correspond to the true object positions up to small shifts related to the encoder and decoder receptive field sizes, the object sizes, and the widths of the Gaussians used in the rendering process. We perform detailed analysis of how the error depends on each of these variables and perform synthetic experiments validating our theoretical predictions up to a precision of individual pixels. We also perform experiments on CLEVR-based data and show that, unlike current SOTA object detection methods (SAM, CutLER), our method's prediction errors always lie within our theoretical bounds. We hope that this work helps open up an avenue of research into object detection methods with theoretical guarantees.

Stanislaw Szymanowicz, Eldar Insafutdinov, Chuanxia Zheng et al.

We propose Flash3D, a method for scene reconstruction and novel view synthesis from a single image which is both very generalisable and efficient. For generalisability, we start from a "foundation" model for monocular depth estimation and extend it to a full 3D shape and appearance reconstructor. For efficiency, we base this extension on feed-forward Gaussian Splatting. Specifically, we predict a first layer of 3D Gaussians at the predicted depth, and then add additional layers of Gaussians that are offset in space, allowing the model to complete the reconstruction behind occlusions and truncations. Flash3D is very efficient, trainable on a single GPU in a day, and thus accessible to most researchers. It achieves state-of-the-art results when trained and tested on RealEstate10k. When transferred to unseen datasets like NYU it outperforms competitors by a large margin. More impressively, when transferred to KITTI, Flash3D achieves better PSNR than methods trained specifically on that dataset. In some instances, it even outperforms recent methods that use multiple views as input. Code, models, demo, and more results are available at https://www.robots.ox.ac.uk/~vgg/research/flash3d/.

Tim Franzmeyer, Aleksandar Shtedritski, Samuel Albanie et al.

Benchmarks have been essential for driving progress in machine learning. A better understanding of LLM capabilities on real world tasks is vital for safe development. Designing adequate LLM benchmarks is challenging: Data from real-world tasks is hard to collect, public availability of static evaluation data results in test data contamination and benchmark overfitting, and periodically generating new evaluation data is tedious and may result in temporally inconsistent results. We introduce HelloFresh, based on continuous streams of real-world data generated by intrinsically motivated human labelers. It covers recent events from X (formerly Twitter) community notes and edits of Wikipedia pages, mitigating the risk of test data contamination and benchmark overfitting. Any X user can propose an X note to add additional context to a misleading post (formerly tweet); if the community classifies it as helpful, it is shown with the post. Similarly, Wikipedia relies on community-based consensus, allowing users to edit articles or revert edits made by other users. Verifying whether an X note is helpful or whether a Wikipedia edit should be accepted are hard tasks that require grounding by querying the web. We backtest state-of-the-art LLMs supplemented with simple web search access and find that HelloFresh yields a temporally consistent ranking. To enable continuous evaluation on HelloFresh, we host a public leaderboard and periodically updated evaluation data at https://tinyurl.com/hello-fresh-LLM.

Dominik A. Kloepfer, João F. Henriques, Dylan Campbell

Extracting point correspondences from two or more views of a scene is a fundamental computer vision problem with particular importance for relative camera pose estimation and structure-from-motion. Existing local feature matching approaches, trained with correspondence supervision on large-scale datasets, obtain highly-accurate matches on the test sets. However, they do not generalise well to new datasets with different characteristics to those they were trained on, unlike classic feature extractors. Instead, they require finetuning, which assumes that ground-truth correspondences or ground-truth camera poses and 3D structure are available. We relax this assumption by removing the requirement of 3D structure, e.g., depth maps or point clouds, and only require camera pose information, which can be obtained from odometry. We do so by replacing correspondence losses with epipolar losses, which encourage putative matches to lie on the associated epipolar line. While weaker than correspondence supervision, we observe that this cue is sufficient for finetuning existing models on new data. We then further relax the assumption of known camera poses by using pose estimates in a novel bootstrapping approach. We evaluate on highly challenging datasets, including an indoor drone dataset and an outdoor smartphone camera dataset, and obtain state-of-the-art results without strong supervision.

Samuel Albanie, Dylan Campbell, João F. Henriques

The field of machine learning has achieved striking progress in recent years, witnessing breakthrough results on language modelling, protein folding and nitpickingly fine-grained dog breed classification. Some even succeeded at playing computer games and board games, a feat both of engineering and of setting their employers' expectations. The central contribution of this work is to carefully examine whether this progress, and technology more broadly, can be expected to continue indefinitely. Through a rigorous application of statistical theory and failure to extrapolate beyond the training data, we answer firmly in the negative and provide details: technology will peak at 3:07 am (BST) on 20th July, 2032. We then explore the implications of this finding, discovering that individuals awake at this ungodly hour with access to a sufficiently powerful computer possess an opportunity for myriad forms of long-term linguistic 'lock in'. All we need is a large (>> 1W) data centre to seize this pivotal moment. By setting our analogue alarm clocks, we propose a tractable algorithm to ensure that, for the future of humanity, the British spelling of colour becomes the default spelling across more than 80% of the global word processing software market.

Jianbo Jiao, João F. Henriques

In this work we investigate how to achieve equivariance to input transformations in deep networks, purely from data, without being given a model of those transformations. Convolutional Neural Networks (CNNs), for example, are equivariant to image translation, a transformation that can be easily modelled (by shifting the pixels vertically or horizontally). Other transformations, such as out-of-plane rotations, do not admit a simple analytic model. We propose an auto-encoder architecture whose embedding obeys an arbitrary set of equivariance relations simultaneously, such as translation, rotation, colour changes, and many others. This means that it can take an input image, and produce versions transformed by a given amount that were not observed before (e.g. a different point of view of the same object, or a colour variation). Despite extending to many (even non-geometric) transformations, our model reduces exactly to a CNN in the special case of translation-equivariance. Equivariances are important for the interpretability and robustness of deep networks, and we demonstrate results of successful re-rendering of transformed versions of input images on several synthetic and real datasets, as well as results on object pose estimation.

Shu Ishida, João F. Henriques

We train embodied neural networks to plan and navigate unseen complex 3D environments, emphasising real-world deployment. Rather than requiring prior knowledge of the agent or environment, the planner learns to model the state transitions and rewards. To avoid the potentially hazardous trial-and-error of reinforcement learning, we focus on differentiable planners such as Value Iteration Networks (VIN), which are trained offline from safe expert demonstrations. Although they work well in small simulations, we address two major limitations that hinder their deployment. First, we observed that current differentiable planners struggle to plan long-term in environments with a high branching complexity. While they should ideally learn to assign low rewards to obstacles to avoid collisions, we posit that the constraints imposed on the network are not strong enough to guarantee the network to learn sufficiently large penalties for every possible collision. We thus impose a structural constraint on the value iteration, which explicitly learns to model any impossible actions. Secondly, we extend the model to work with a limited perspective camera under translation and rotation, which is crucial for real robot deployment. Many VIN-like planners assume a 360 degrees or overhead view without rotation. In contrast, our method uses a memory-efficient lattice map to aggregate CNN embeddings of partial observations, and models the rotational dynamics explicitly using a 3D state-space grid (translation and rotation). Our proposals significantly improve semantic navigation and exploration on several 2D and 3D environments, succeeding in settings that are otherwise challenging for this class of methods. As far as we know, we are the first to successfully perform differentiable planning on the difficult Active Vision Dataset, consisting of real images captured from a robot.

Tim Franzmeyer, Mateusz Malinowski, João F. Henriques

Can artificial agents learn to assist others in achieving their goals without knowing what those goals are? Generic reinforcement learning agents could be trained to behave altruistically towards others by rewarding them for altruistic behaviour, i.e., rewarding them for benefiting other agents in a given situation. Such an approach assumes that other agents' goals are known so that the altruistic agent can cooperate in achieving those goals. However, explicit knowledge of other agents' goals is often difficult to acquire. In the case of human agents, their goals and preferences may be difficult to express fully; they might be ambiguous or even contradictory. Thus, it is beneficial to develop agents that do not depend on external supervision and learn altruistic behaviour in a task-agnostic manner. We propose to act altruistically towards other agents by giving them more choice and allowing them to achieve their goals better. Some concrete examples include opening a door for others or safeguarding them to pursue their objectives without interference. We formalize this concept and propose an altruistic agent that learns to increase the choices another agent has by preferring to maximize the number of states that the other agent can reach in its future. We evaluate our approach in three different multi-agent environments where another agent's success depends on altruistic behaviour. Finally, we show that our unsupervised agents can perform comparably to agents explicitly trained to work cooperatively, in some cases even outperforming them.

Andreea-Maria Oncescu, A. Sophia Koepke, João F. Henriques et al.

We consider the task of retrieving audio using free-form natural language queries. To study this problem, which has received limited attention in the existing literature, we introduce challenging new benchmarks for text-based audio retrieval using text annotations sourced from the Audiocaps and Clotho datasets. We then employ these benchmarks to establish baselines for cross-modal audio retrieval, where we demonstrate the benefits of pre-training on diverse audio tasks. We hope that our benchmarks will inspire further research into cross-modal text-based audio retrieval with free-form text queries.

Andreea-Maria Oncescu, João F. Henriques, Yang Liu et al.

We introduce QuerYD, a new large-scale dataset for retrieval and event localisation in video. A unique feature of our dataset is the availability of two audio tracks for each video: the original audio, and a high-quality spoken description of the visual content. The dataset is based on YouDescribe, a volunteer project that assists visually-impaired people by attaching voiced narrations to existing YouTube videos. This ever-growing collection of videos contains highly detailed, temporally aligned audio and text annotations. The content descriptions are more relevant than dialogue, and more detailed than previous description attempts, which can be observed to contain many superficial or uninformative descriptions. To demonstrate the utility of the QuerYD dataset, we show that it can be used to train and benchmark strong models for retrieval and event localisation. Data, code and models are made publicly available, and we hope that QuerYD inspires further research on video understanding with written and spoken natural language.

Mandela Patrick, Yuki M. Asano, Polina Kuznetsova et al.

In the image domain, excellent representations can be learned by inducing invariance to content-preserving transformations via noise contrastive learning. In this paper, we generalize contrastive learning to a wider set of transformations, and their compositions, for which either invariance or distinctiveness is sought. We show that it is not immediately obvious how existing methods such as SimCLR can be extended to do so. Instead, we introduce a number of formal requirements that all contrastive formulations must satisfy, and propose a practical construction which satisfies these requirements. In order to maximise the reach of this analysis, we express all components of noise contrastive formulations as the choice of certain generalized transformations of the data (GDTs), including data sampling. We then consider videos as an example of data in which a large variety of transformations are applicable, accounting for the extra modalities -- for which we analyze audio and text -- and the dimension of time. We find that being invariant to certain transformations and distinctive to others is critical to learning effective video representations, improving the state-of-the-art for multiple benchmarks by a large margin, and even surpassing supervised pretraining.

Luca Bertinetto, João F. Henriques, Philip H. S. Torr et al.

Adapting deep networks to new concepts from a few examples is challenging, due to the high computational requirements of standard fine-tuning procedures. Most work on few-shot learning has thus focused on simple learning techniques for adaptation, such as nearest neighbours or gradient descent. Nonetheless, the machine learning literature contains a wealth of methods that learn non-deep models very efficiently. In this paper, we propose to use these fast convergent methods as the main adaptation mechanism for few-shot learning. The main idea is to teach a deep network to use standard machine learning tools, such as ridge regression, as part of its own internal model, enabling it to quickly adapt to novel data. This requires back-propagating errors through the solver steps. While normally the cost of the matrix operations involved in such a process would be significant, by using the Woodbury identity we can make the small number of examples work to our advantage. We propose both closed-form and iterative solvers, based on ridge regression and logistic regression components. Our methods constitute a simple and novel approach to the problem of few-shot learning and achieve performance competitive with or superior to the state of the art on three benchmarks.

Jack Valmadre, Luca Bertinetto, João F. Henriques et al.

We introduce the OxUvA dataset and benchmark for evaluating single-object tracking algorithms. Benchmarks have enabled great strides in the field of object tracking by defining standardized evaluations on large sets of diverse videos. However, these works have focused exclusively on sequences that are just tens of seconds in length and in which the target is always visible. Consequently, most researchers have designed methods tailored to this "short-term" scenario, which is poorly representative of practitioners' needs. Aiming to address this disparity, we compile a long-term, large-scale tracking dataset of sequences with average length greater than two minutes and with frequent target object disappearance. The OxUvA dataset is much larger than the object tracking datasets of recent years: it comprises 366 sequences spanning 14 hours of video. We assess the performance of several algorithms, considering both the ability to locate the target and to determine whether it is present or absent. Our goal is to offer the community a large and diverse benchmark to enable the design and evaluation of tracking methods ready to be used "in the wild". The project website is http://oxuva.net

Jack Valmadre, Luca Bertinetto, João F. Henriques et al.

The Correlation Filter is an algorithm that trains a linear template to discriminate between images and their translations. It is well suited to object tracking because its formulation in the Fourier domain provides a fast solution, enabling the detector to be re-trained once per frame. Previous works that use the Correlation Filter, however, have adopted features that were either manually designed or trained for a different task. This work is the first to overcome this limitation by interpreting the Correlation Filter learner, which has a closed-form solution, as a differentiable layer in a deep neural network. This enables learning deep features that are tightly coupled to the Correlation Filter. Experiments illustrate that our method has the important practical benefit of allowing lightweight architectures to achieve state-of-the-art performance at high framerates.

Samuel Albanie, Sébastien Ehrhardt, João F. Henriques

While the costs of human violence have attracted a great deal of attention from the research community, the effects of the network-on-network (NoN) violence popularised by Generative Adversarial Networks have yet to be addressed. In this work, we quantify the financial, social, spiritual, cultural, grammatical and dermatological impact of this aggression and address the issue by proposing a more peaceful approach which we term Generative Unadversarial Networks (GUNs). Under this framework, we simultaneously train two models: a generator G that does its best to capture whichever data distribution it feels it can manage, and a motivator M that helps G to achieve its dream. Fighting is strictly verboten and both models evolve by learning to respect their differences. The framework is both theoretically and electrically grounded in game theory, and can be viewed as a winner-shares-all two-player game in which both players work as a team to achieve the best score. Experiments show that by working in harmony, the proposed model is able to claim both the moral and log-likelihood high ground. Our work builds on a rich history of carefully argued position-papers, published as anonymous YouTube comments, which prove that the optimal solution to NoN violence is more GUNs.

João F. Henriques, Andrea Vedaldi

Convolutional Neural Networks (CNNs) are extremely efficient, since they exploit the inherent translation-invariance of natural images. However, translation is just one of a myriad of useful spatial transformations. Can the same efficiency be attained when considering other spatial invariances? Such generalized convolutions have been considered in the past, but at a high computational cost. We present a construction that is simple and exact, yet has the same computational complexity that standard convolutions enjoy. It consists of a constant image warp followed by a simple convolution, which are standard blocks in deep learning toolboxes. With a carefully crafted warp, the resulting architecture can be made equivariant to a wide range of two-parameter spatial transformations. We show encouraging results in realistic scenarios, including the estimation of vehicle poses in the Google Earth dataset (rotation and scale), and face poses in Annotated Facial Landmarks in the Wild (3D rotations under perspective).

Luca Bertinetto, Jack Valmadre, João F. Henriques et al.

The problem of arbitrary object tracking has traditionally been tackled by learning a model of the object's appearance exclusively online, using as sole training data the video itself. Despite the success of these methods, their online-only approach inherently limits the richness of the model they can learn. Recently, several attempts have been made to exploit the expressive power of deep convolutional networks. However, when the object to track is not known beforehand, it is necessary to perform Stochastic Gradient Descent online to adapt the weights of the network, severely compromising the speed of the system. In this paper we equip a basic tracking algorithm with a novel fully-convolutional Siamese network trained end-to-end on the ILSVRC15 dataset for object detection in video. Our tracker operates at frame-rates beyond real-time and, despite its extreme simplicity, achieves state-of-the-art performance in multiple benchmarks.

Luca Bertinetto, João F. Henriques, Jack Valmadre et al.

One-shot learning is usually tackled by using generative models or discriminative embeddings. Discriminative methods based on deep learning, which are very effective in other learning scenarios, are ill-suited for one-shot learning as they need large amounts of training data. In this paper, we propose a method to learn the parameters of a deep model in one shot. We construct the learner as a second deep network, called a learnet, which predicts the parameters of a pupil network from a single exemplar. In this manner we obtain an efficient feed-forward one-shot learner, trained end-to-end by minimizing a one-shot classification objective in a learning to learn formulation. In order to make the construction feasible, we propose a number of factorizations of the parameters of the pupil network. We demonstrate encouraging results by learning characters from single exemplars in Omniglot, and by tracking visual objects from a single initial exemplar in the Visual Object Tracking benchmark.