Dylan Campbell

Ge-Peng Ji, Jing Zhang, Dylan Campbell et al.

Unlike existing fully-supervised approaches, we rethink colorectal polyp segmentation from an out-of-distribution perspective with a simple but effective self-supervised learning approach. We leverage the ability of masked autoencoders -- self-supervised vision transformers trained on a reconstruction task -- to learn in-distribution representations; here, the distribution of healthy colon images. We then perform out-of-distribution reconstruction and inference, with feature space standardisation to align the latent distribution of the diverse abnormal samples with the statistics of the healthy samples. We generate per-pixel anomaly scores for each image by calculating the difference between the input and reconstructed images and use this signal for out-of-distribution (ie, polyp) segmentation. Experimental results on six benchmarks show that our model has excellent segmentation performance and generalises across datasets. Our code is publicly available at https://github.com/GewelsJI/Polyp-OOD.

Frederic Z. Zhang, Yuhui Yuan, Dylan Campbell et al.

Recently, the DETR framework has emerged as the dominant approach for human--object interaction (HOI) research. In particular, two-stage transformer-based HOI detectors are amongst the most performant and training-efficient approaches. However, these often condition HOI classification on object features that lack fine-grained contextual information, eschewing pose and orientation information in favour of visual cues about object identity and box extremities. This naturally hinders the recognition of complex or ambiguous interactions. In this work, we study these issues through visualisations and carefully designed experiments. Accordingly, we investigate how best to re-introduce image features via cross-attention. With an improved query design, extensive exploration of keys and values, and box pair positional embeddings as spatial guidance, our model with enhanced predicate visual context (PViC) outperforms state-of-the-art methods on the HICO-DET and V-COCO benchmarks, while maintaining low training cost.

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.

Yujiao Shi, Xin Yu, Liu Liu et al.

We address the problem of ground-to-satellite image geo-localization, that is, estimating the camera latitude, longitude and orientation (azimuth angle) by matching a query image captured at the ground level against a large-scale database with geotagged satellite images. Our prior arts treat the above task as pure image retrieval by selecting the most similar satellite reference image matching the ground-level query image. However, such an approach often produces coarse location estimates because the geotag of the retrieved satellite image only corresponds to the image center while the ground camera can be located at any point within the image. To further consolidate our prior research findings, we present a novel geometry-aware geo-localization method. Our new method is able to achieve the fine-grained location of a query image, up to pixel size precision of the satellite image, once its coarse location and orientation have been determined. Moreover, we propose a new geometry-aware image retrieval pipeline to improve the coarse localization accuracy. Apart from a polar transform in our conference work, this new pipeline also maps satellite image pixels to the ground-level plane in the ground-view via a geometry-constrained projective transform to emphasize informative regions, such as road structures, for cross-view geo-localization. Extensive quantitative and qualitative experiments demonstrate the effectiveness of our newly proposed framework. We also significantly improve the performance of coarse localization results compared to the state-of-the-art in terms of location recalls.

Peter Tu, Zhaoyuan Yang, Richard Hartley et al.

This paper begins with a description of methods for estimating image probability density functions that reflects the observation that such data is usually constrained to lie in restricted regions of the high-dimensional image space-not every pattern of pixels is an image. It is common to say that images lie on a lower-dimensional manifold in the high-dimensional space. However, it is not the case that all points on the manifold have an equal probability of being images. Images are unevenly distributed on the manifold, and our task is to devise ways to model this distribution as a probability distribution. We therefore consider popular generative models. For our purposes, generative/probabilistic models should have the properties of 1) sample generation: the possibility to sample from this distribution with the modelled density function, and 2) probability computation: given a previously unseen sample from the dataset of interest, one should be able to compute its probability, at least up to a normalising constant. To this end, we investigate the use of methods such as normalising flow and diffusion models. We then show how semantic interpretations are used to describe points on the manifold. To achieve this, we consider an emergent language framework that uses variational encoders for a disentangled representation of points that reside on a given manifold. Trajectories between points on a manifold can then be described as evolving semantic descriptions. We also show that such probabilistic descriptions (bounded) can be used to improve semantic consistency by constructing defences against adversarial attacks. We evaluate our methods with improved semantic robustness and OoD detection capability, explainable and editable semantic interpolation, and improved classification accuracy under patch attacks. We also discuss the limitation in diffusion models.

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.

Zhaoyuan Yang, Zhengyang Yu, Zhiwei Xu et al.

We present a diffusion-based image morphing approach with perceptually-uniform sampling (IMPUS) that produces smooth, direct and realistic interpolations given an image pair. The embeddings of two images may lie on distinct conditioned distributions of a latent diffusion model, especially when they have significant semantic difference. To bridge this gap, we interpolate in the locally linear and continuous text embedding space and Gaussian latent space. We first optimize the endpoint text embeddings and then map the images to the latent space using a probability flow ODE. Unlike existing work that takes an indirect morphing path, we show that the model adaptation yields a direct path and suppresses ghosting artifacts in the interpolated images. To achieve this, we propose a heuristic bottleneck constraint based on a novel relative perceptual path diversity score that automatically controls the bottleneck size and balances the diversity along the path with its directness. We also propose a perceptually-uniform sampling technique that enables visually smooth changes between the interpolated images. Extensive experiments validate that our IMPUS can achieve smooth, direct, and realistic image morphing and is adaptable to several other generative tasks.

Marius Kästingschäfer, Théo Gieruc, Sebastian Bernhard et al.

Models for egocentric 3D and 4D reconstruction, including few-shot interpolation and extrapolation settings, can benefit from having images from exocentric viewpoints as supervision signals. No existing dataset provides the necessary mixture of complex, dynamic, and multi-view data. To facilitate the development of 3D and 4D reconstruction methods in the autonomous driving context, we propose a Synthetic Ego--Exo Dynamic 4D (SEED4D) data generator and dataset. We present a customizable, easy-to-use data generator for spatio-temporal multi-view data creation. Our open-source data generator allows the creation of synthetic data for camera setups commonly used in the NuScenes, KITTI360, and Waymo datasets. Additionally, SEED4D encompasses two large-scale multi-view synthetic urban scene datasets. Our static (3D) dataset encompasses 212k inward- and outward-facing vehicle images from 2k scenes, while our dynamic (4D) dataset contains 16.8M images from 10k trajectories, each sampled at 100 points in time with egocentric images, exocentric images, and LiDAR data. The datasets and the data generator can be found at https://seed4d.github.io/.

Qingtao Yu, Changlin Song, Minghao Sun et al.

A prominent approach to test-time scaling for text-to-image diffusion models formulates the problem as a search over multiple noise seeds, selecting the one that maximizes a certain image-reward function. The effectiveness of this strategy heavily depends on the number and diversity of noise seeds explored. However, verifying each candidate is computationally expensive, because each must be fully denoised before a reward can be computed. This severely limits the number of samples that can be explored under a fixed budget. We propose test-time scaling with noise-aware pruning (TTSnap), a framework that prunes low-quality candidates without fully denoising them. The key challenge is that reward models are learned in the clean image domain, and the ranking of rewards predicted for intermediate estimates are often inconsistent with those predicted for clean images. To overcome this, we train noise-aware reward models via self-distillation to align the reward for intermediate estimates with that of the final clean images. To stabilize learning across different noise levels, we adopt a curriculum training strategy that progressively shifts the data domain from clean images to noise images. In addition, we introduce a new metric that measures reward alignment and computational budget utilization. Experiments demonstrate that our approach improves performance by over 16\% compared with existing methods, enabling more efficient and effective test-time scaling. It also provides orthogonal gains when combined with post-training techniques and local test-time optimization. Code: https://github.com/TerrysLearning/TTSnap/.

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

Shihao Jiang, Dylan Campbell, Yao Lu et al.

Occlusions pose a significant challenge to optical flow algorithms that rely on local evidences. We consider an occluded point to be one that is imaged in the first frame but not in the next, a slight overloading of the standard definition since it also includes points that move out-of-frame. Estimating the motion of these points is extremely difficult, particularly in the two-frame setting. Previous work relies on CNNs to learn occlusions, without much success, or requires multiple frames to reason about occlusions using temporal smoothness. In this paper, we argue that the occlusion problem can be better solved in the two-frame case by modelling image self-similarities. We introduce a global motion aggregation module, a transformer-based approach to find long-range dependencies between pixels in the first image, and perform global aggregation on the corresponding motion features. We demonstrate that the optical flow estimates in the occluded regions can be significantly improved without damaging the performance in non-occluded regions. This approach obtains new state-of-the-art results on the challenging Sintel dataset, improving the average end-point error by 13.6% on Sintel Final and 13.7% on Sintel Clean. At the time of submission, our method ranks first on these benchmarks among all published and unpublished approaches. Code is available at https://github.com/zacjiang/GMA

Subhransu S. Bhattacharjee, Dylan Campbell, Rahul Shome

A single egocentric image typically captures only a small portion of the floor, yet a complete metric traversability map of the surroundings would better serve applications such as indoor navigation. We introduce FlatLands, a dataset and benchmark for single-view bird's-eye view (BEV) floor completion. The dataset contains 270,575 observations from 17,656 real metric indoor scenes drawn from six existing datasets, with aligned observation, visibility, validity, and ground-truth BEV maps, and the benchmark includes both in- and out-of-distribution evaluation protocols. We compare training-free approaches, deterministic models, ensembles, and stochastic generative models. Finally, we instantiate the task as an end-to-end monocular RGB-to-floormaps pipeline. FlatLands provides a rigorous testbed for uncertainty-aware indoor mapping and generative completion for embodied navigation.

Weijie Tu, Weijian Deng, Dylan Campbell et al.

Vision-Language Models (VLMs) have emerged as the dominant approach for zero-shot recognition, adept at handling diverse scenarios and significant distribution changes. However, their deployment in risk-sensitive areas requires a deeper understanding of their uncertainty estimation capabilities, a relatively uncharted area. In this study, we explore the calibration properties of VLMs across different architectures, datasets, and training strategies. In particular, we analyze the uncertainty estimation performance of VLMs when calibrated in one domain, label set or hierarchy level, and tested in a different one. Our findings reveal that while VLMs are not inherently calibrated for uncertainty, temperature scaling significantly and consistently improves calibration, even across shifts in distribution and changes in label set. Moreover, VLMs can be calibrated with a very small set of examples. Through detailed experimentation, we highlight the potential applications and importance of our insights, aiming for more reliable and effective use of VLMs in critical, real-world scenarios.

Qingtao Yu, Jaskirat Singh, Zhaoyuan Yang et al.

Diffusion models indirectly estimate the probability density over a data space, which can be used to study its structure. In this work, we show that geodesics can be computed in diffusion latent space, where the norm induced by the spatially-varying inner product is inversely proportional to the probability density. In this formulation, a path that traverses a high density (that is, probable) region of image latent space is shorter than the equivalent path through a low density region. We present algorithms for solving the associated initial and boundary value problems and show how to compute the probability density along the path and the geodesic distance between two points. Using these techniques, we analyze how closely video clips approximate geodesics in a pre-trained image diffusion space. Finally, we demonstrate how these techniques can be applied to training-free image sequence interpolation and extrapolation, given a pre-trained image diffusion model.

Zhenyue Qin, Yiqun Zhang, Yang Liu et al.

Generative text-to-image models, such as Stable Diffusion, have demonstrated a remarkable ability to generate diverse, high-quality images. However, they are surprisingly inept when it comes to rendering human hands, which are often anatomically incorrect or reside in the "uncanny valley". In this paper, we propose a method HandCraft for restoring such malformed hands. This is achieved by automatically constructing masks and depth images for hands as conditioning signals using a parametric model, allowing a diffusion-based image editor to fix the hand's anatomy and adjust its pose while seamlessly integrating the changes into the original image, preserving pose, color, and style. Our plug-and-play hand restoration solution is compatible with existing pretrained diffusion models, and the restoration process facilitates adoption by eschewing any fine-tuning or training requirements for the diffusion models. We also contribute MalHand datasets that contain generated images with a wide variety of malformed hands in several styles for hand detector training and hand restoration benchmarking, and demonstrate through qualitative and quantitative evaluation that HandCraft not only restores anatomical correctness but also maintains the integrity of the overall image.

Weijie Tu, Weijian Deng, Dylan Campbell et al.

Can the relative performance of a pre-trained large multimodal model (LMM) be predicted without access to labels? As LMMs proliferate, it becomes increasingly important to develop efficient ways to choose between them when faced with new data or tasks. The usual approach does the equivalent of giving the models an exam and marking them. We opt to avoid marking and the associated labor of determining the ground-truth answers. Instead, we explore other signals elicited and ascertain how well the models know their own limits, evaluating the effectiveness of these signals at unsupervised model ranking. We evaluate $47$ state-of-the-art LMMs (\eg, LLaVA) across $9$ visual question answering benchmarks, analyzing how well uncertainty-based metrics can predict relative model performance. Our findings show that uncertainty scores derived from softmax distributions provide a robust and consistent basis for ranking models across various tasks. This facilitates the ranking of LMMs on unlabeled data, providing a practical approach for selecting models for diverse target domains without requiring manual annotation.

Yiqun Zhang, Zhenyue Qin, Yang Liu et al.

We introduce a pipeline to address anatomical inaccuracies in Stable Diffusion generated hand images. The initial step involves constructing a specialized dataset, focusing on hand anomalies, to train our models effectively. A finetuned detection model is pivotal for precise identification of these anomalies, ensuring targeted correction. Body pose estimation aids in understanding hand orientation and positioning, crucial for accurate anomaly correction. The integration of ControlNet and InstructPix2Pix facilitates sophisticated inpainting and pixel-level transformation, respectively. This dual approach allows for high-fidelity image adjustments. This comprehensive approach ensures the generation of images with anatomically accurate hands, closely resembling real-world appearances. Our experimental results demonstrate the pipeline's efficacy in enhancing hand image realism in Stable Diffusion outputs. We provide an online demo at https://fixhand.yiqun.io

Yue Yin, Enze Tao, Dylan Campbell

Generative image models can produce convincingly real images, with plausible shapes, textures, layouts and lighting. However, one domain in which they perform notably poorly is in the synthesis of transparent objects, which exhibit refraction, reflection, absorption and scattering. Refraction is a particular challenge, because refracted pixel rays often intersect with surfaces observed in other parts of the image, providing a constraint on the color. It is clear from inspection that generative models have not distilled the laws of optics sufficiently well to accurately render refractive objects. In this work, we consider the problem of generating images with accurate refraction, given a text prompt. We synchronize the pixels within the object's boundary with those outside by warping and merging the pixels using Snell's Law of Refraction, at each step of the generation trajectory. For those surfaces that are not directly observed in the image, but are visible via refraction or reflection, we recover their appearance by synchronizing the image with a second generated image -- a panorama centered at the object -- using the same warping and merging procedure. We demonstrate that our approach generates much more optically-plausible images that respect the physical constraints.

Sam Bahrami, Dylan Campbell

Modern scene reconstruction methods are able to accurately recover 3D surfaces that are visible in one or more images. However, this leads to incomplete reconstructions, missing all occluded surfaces. While much progress has been made on reconstructing entire objects given partial observations using generative models, the structural elements of a scene, like the walls, floors and ceilings, have received less attention. We argue that these scene elements should be relatively easy to predict, since they are typically planar, repetitive and simple, and so less costly approaches may be suitable. In this work, we present a synthetic dataset -- Room Envelopes -- that facilitates progress on this task by providing a set of RGB images and two associated pointmaps for each image: one capturing the visible surface and one capturing the first surface once fittings and fixtures are removed, that is, the structural layout. As we show, this enables direct supervision for feed-forward monocular geometry estimators that predict both the first visible surface and the first layout surface. This confers an understanding of the scene's extent, as well as the shape and location of its objects.

Daniel Wang, Evan Markou, Dylan Campbell

While backpropagation--reverse-mode automatic differentiation--has been extraordinarily successful in deep learning, it requires two passes (forward and backward) through the neural network and the storage of intermediate activations. Existing gradient estimation methods that instead use forward-mode automatic differentiation struggle to scale beyond small networks due to the high variance of the estimates. Efforts to mitigate this have so far introduced significant bias to the estimates, reducing their utility. We introduce a gradient estimation approach that reduces both bias and variance by manipulating upstream Jacobian matrices when computing guess directions. It shows promising results and has the potential to scale to larger networks, indeed performing better as the network width is increased. Our understanding of this method is facilitated by analyses of bias and variance, and their connection to the low-dimensional structure of neural network gradients.

Subhransu S. Bhattacharjee, Hao Lu, Dylan Campbell et al.

Priors are vital for planning under partial observability, yet difficult to obtain in practice. We present a sampling-based pipeline that leverages large-scale pretrained generative models to produce probabilistic priors capturing environmental uncertainty and spatio-semantic relationships in a zero-shot manner. Conditioned on partial observations, the pipeline recovers complete RGB-D point cloud samples with occupancy and target semantics, formulated to be directly useful in configuration-space planning. We establish a Matterport3D benchmark of rooms partially visible through doorways, where a robot must navigate to an unobserved target object. Effective priors for this setting must represent both occupancy and target-location uncertainty in unobserved regions. Experiments show that our approach recovers commonsense spatial semantics consistent with ground truth, yielding diverse, clean 3D point clouds usable in motion planning, highlight the promise of generative models as a rich source of priors for robotic planning.

Changlin Song, Yunzhong Hou, Michael Randall Barnes et al.

Extreme amodal detection is the task of inferring the 2D location of objects that are not fully visible in the input image but are visible within an expanded field-of-view. This differs from amodal detection, where the object is partially visible within the input image, but is occluded. In this paper, we consider the sub-problem of face detection, since this class provides motivating applications involving safety and privacy, but do not tailor our method specifically to this class. Existing approaches rely on image sequences so that missing detections may be interpolated from surrounding frames or make use of generative models to sample possible completions. In contrast, we consider the single-image task and propose a more efficient, sample-free approach that makes use of the contextual cues from the image to infer the presence of unseen faces. We design a heatmap-based extreme amodal object detector that addresses the problem of efficiently predicting a lot (the out-of-frame region) from a little (the image) with a selective coarse-to-fine decoder. Our method establishes strong results for this new task, even outperforming less efficient generative approaches.

Zhenyue Qin, Yang Liu, Yu Yin et al.

Vision-threatening eye diseases pose a major global health burden, with timely diagnosis limited by workforce shortages and restricted access to specialized care. While multimodal large language models (MLLMs) show promise for medical image interpretation, advancing MLLMs for ophthalmology is hindered by the lack of comprehensive benchmark datasets suitable for evaluating generative models. We present a large-scale multimodal ophthalmology benchmark comprising 32,633 instances with multi-granular annotations across 12 common ophthalmic conditions and 5 imaging modalities. The dataset integrates imaging, anatomical structures, demographics, and free-text annotations, supporting anatomical structure recognition, disease screening, disease staging, and demographic prediction for bias evaluation. This work extends our preliminary LMOD benchmark with three major enhancements: (1) nearly 50% dataset expansion with substantial enlargement of color fundus photography; (2) broadened task coverage including binary disease diagnosis, multi-class diagnosis, severity classification with international grading standards, and demographic prediction; and (3) systematic evaluation of 24 state-of-the-art MLLMs. Our evaluations reveal both promise and limitations. Top-performing models achieved ~58% accuracy in disease screening under zero-shot settings, and performance remained suboptimal for challenging tasks like disease staging. We will publicly release the dataset, curation pipeline, and leaderboard to potentially advance ophthalmic AI applications and reduce the global burden of vision-threatening diseases.

Yumin Li, Dylan Campbell

Estimating metric relative camera pose from a pair of images is of great importance for 3D reconstruction and localisation. However, conventional two-view pose estimation methods are not metric, with camera translation known only up to a scale, and struggle with wide baselines and textureless or reflective surfaces. This paper introduces GARPS, a training-free framework that casts this problem as the direct alignment of two independently reconstructed 3D scenes. GARPS leverages a metric monocular depth estimator and a Gaussian scene reconstructor to obtain a metric 3D Gaussian Mixture Model (GMM) for each image. It then refines an initial pose from a feed-forward two-view pose estimator by optimising a differentiable GMM alignment objective. This objective jointly considers geometric structure, view-independent colour, anisotropic covariance, and semantic feature consistency, and is robust to occlusions and texture-poor regions without requiring explicit 2D correspondences. Extensive experiments on the Real\-Estate10K dataset demonstrate that GARPS outperforms both classical and state-of-the-art learning-based methods, including MASt3R. These results highlight the potential of bridging single-view perception with multi-view geometry to achieve robust and metric relative pose estimation.

Sam Bahrami, Dylan Campbell

Feed-forward 3D reconstruction methods aim to predict the 3D structure of a scene directly from input images, providing a faster alternative to per-scene optimization approaches. Significant progress has been made in single-view and few-view reconstruction using learned priors that infer object shape and appearance, even for unobserved regions. However, there is substantial potential to enhance these methods by better leveraging information from multiple views when available. To address this, we propose a few-view reconstruction model that more effectively harnesses multi-view information. Our approach introduces a simple mechanism that connects the 3D representation with pixel rays from the input views, allowing for preferential sharing of information between nearby 3D locations and between 3D locations and nearby pixel rays. We achieve this by defining the 3D representation as a set of structured, feature-augmented lines; the PlückeRF representation. Using this representation, we demonstrate improvements in reconstruction quality over the equivalent triplane representation and state-of-the-art feedforward reconstruction methods.

Yue Yin, Enze Tao, Weijian Deng et al.

Modern 3D reconstruction and novel view synthesis approaches have demonstrated strong performance on scenes with opaque Lambertian objects. However, most assume straight light paths and therefore cannot properly handle refractive and reflective materials. Moreover, datasets specialized for these effects are limited, stymieing efforts to evaluate performance and develop suitable techniques. In this work, we introduce a synthetic RefRef dataset and benchmark for reconstructing scenes with refractive and reflective objects from posed images. Our dataset has 50 such objects of varying complexity, from single-material convex shapes to multi-material non-convex shapes, each placed in three different background types, resulting in 150 scenes. We also propose an oracle method that, given the object geometry and refractive indices, calculates accurate light paths for neural rendering, and an approach based on this that avoids these assumptions. We benchmark these against several state-of-the-art methods and show that all methods lag significantly behind the oracle, highlighting the challenges of the task and dataset.

Mo Zhou, Jianwei Wang, Xuanmeng Zhang et al.

This paper studies optical flow estimation, a critical task in motion analysis with applications in autonomous navigation, action recognition, and film production. Traditional optical flow methods require consecutive frames, which are often unavailable due to limitations in data acquisition or real-world scene disruptions. Thus, single-frame optical flow estimation is emerging in the literature. However, existing single-frame approaches suffer from two major limitations: (1) they rely on labeled training data, making them task-specific, and (2) they produce deterministic predictions, failing to capture motion uncertainty. To overcome these challenges, we propose ProbDiffFlow, a training-free framework that estimates optical flow distributions from a single image. Instead of directly predicting motion, ProbDiffFlow follows an estimation-by-synthesis paradigm: it first generates diverse plausible future frames using a diffusion-based model, then estimates motion from these synthesized samples using a pre-trained optical flow model, and finally aggregates the results into a probabilistic flow distribution. This design eliminates the need for task-specific training while capturing multiple plausible motions. Experiments on both synthetic and real-world datasets demonstrate that ProbDiffFlow achieves superior accuracy, diversity, and efficiency, outperforming existing single-image and two-frame baselines.

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/.

Joao F. Henriques, Dylan Campbell, Tengda Han

Two years ago, Stable Diffusion achieved super-human performance at generating images with super-human numbers of fingers. Following the steady decline of its technical novelty, we propose Stale Diffusion, a method that solidifies and ossifies Stable Diffusion in a maximum-entropy state. Stable Diffusion works analogously to a barn (the Stable) from which an infinite set of horses have escaped (the Diffusion). As the horses have long left the barn, our proposal may be seen as antiquated and irrelevant. Nevertheless, we vigorously defend our claim of novelty by identifying as early adopters of the Slow Science Movement, which will produce extremely important pearls of wisdom in the future. Our speed of contributions can also be seen as a quasi-static implementation of the recent call to pause AI experiments, which we wholeheartedly support. As a result of a careful archaeological expedition to 18-months-old Git commit histories, we found that naturally-accumulating errors have produced a novel entropy-maximising Stale Diffusion method, that can produce sleep-inducing hyper-realistic 5D video that is as good as one's imagination.

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.

Stephen Gould, Dylan Campbell, Itzik Ben-Shabat et al.

Deep declarative networks and other recent related works have shown how to differentiate the solution map of a (continuous) parametrized optimization problem, opening up the possibility of embedding mathematical optimization problems into end-to-end learnable models. These differentiability results can lead to significant memory savings by providing an expression for computing the derivative without needing to unroll the steps of the forward-pass optimization procedure during the backward pass. However, the results typically require inverting a large Hessian matrix, which is computationally expensive when implemented naively. In this work we study two applications of deep declarative networks -- robust vector pooling and optimal transport -- and show how problem structure can be exploited to obtain very efficient backward pass computations in terms of both time and memory. Our ideas can be used as a guide for improving the computational performance of other novel deep declarative nodes.

Frederic Z. Zhang, Dylan Campbell, Stephen Gould

Recent developments in transformer models for visual data have led to significant improvements in recognition and detection tasks. In particular, using learnable queries in place of region proposals has given rise to a new class of one-stage detection models, spearheaded by the Detection Transformer (DETR). Variations on this one-stage approach have since dominated human-object interaction (HOI) detection. However, the success of such one-stage HOI detectors can largely be attributed to the representation power of transformers. We discovered that when equipped with the same transformer, their two-stage counterparts can be more performant and memory-efficient, while taking a fraction of the time to train. In this work, we propose the Unary-Pairwise Transformer, a two-stage detector that exploits unary and pairwise representations for HOIs. We observe that the unary and pairwise parts of our transformer network specialise, with the former preferentially increasing the scores of positive examples and the latter decreasing the scores of negative examples. We evaluate our method on the HICO-DET and V-COCO datasets, and significantly outperform state-of-the-art approaches. At inference time, our model with ResNet50 approaches real-time performance on a single GPU.

Ali Cheraghian, Shafinn Rahman, Townim F. Chowdhury et al.

Zero-shot learning, the task of learning to recognize new classes not seen during training, has received considerable attention in the case of 2D image classification. However, despite the increasing ubiquity of 3D sensors, the corresponding 3D point cloud classification problem has not been meaningfully explored and introduces new challenges. In this paper, we identify some of the challenges and apply 2D Zero-Shot Learning (ZSL) methods in the 3D domain to analyze the performance of existing models. Then, we propose a novel approach to address the issues specific to 3D ZSL. We first present an inductive ZSL process and then extend it to the transductive ZSL and Generalized ZSL (GZSL) settings for 3D point cloud classification. To this end, a novel loss function is developed that simultaneously aligns seen semantics with point cloud features and takes advantage of unlabeled test data to address some known issues (e.g., the problems of domain adaptation, hubness, and data bias). While designed for the particularities of 3D point cloud classification, the method is shown to also be applicable to the more common use-case of 2D image classification. An extensive set of experiments is carried out, establishing state-of-the-art for ZSL and GZSL on synthetic (ModelNet40, ModelNet10, McGill) and real (ScanObjectNN) 3D point cloud datasets.

Yujiao Shi, Dylan Campbell, Xin Yu et al.

This paper presents a new approach for synthesizing a novel street-view panorama given an overhead satellite image. Taking a small satellite image patch as input, our method generates a Google's omnidirectional street-view type panorama, as if it is captured from the same geographical location as the center of the satellite patch. Existing works tackle this task as an image generation problem which adopts generative adversarial networks to implicitly learn the cross-view transformations, while ignoring the domain relevance. In this paper, we propose to explicitly establish the geometric correspondences between the two-view images so as to facilitate the cross-view transformation learning. Specifically, we observe that when a 3D point in the real world is visible in both views, there is a deterministic mapping between the projected points in the two-view images given the height information of this 3D point. Motivated by this, we develop a novel Satellite to Street-view image Projection (S2SP) module which explicitly establishes such geometric correspondences and projects the satellite images to the street viewpoint. With these projected satellite images as network input, we next employ a generator to synthesize realistic street-view panoramas that are geometrically consistent with the satellite images. Our S2SP module is differentiable and the whole framework is trained in an end-to-end manner. Extensive experimental results on two cross-view benchmark datasets demonstrate that our method generates images that better respect the scene geometry than existing approaches.

Frederic Z. Zhang, Dylan Campbell, Stephen Gould

We address the problem of detecting human-object interactions in images using graphical neural networks. Unlike conventional methods, where nodes send scaled but otherwise identical messages to each of their neighbours, we propose to condition messages between pairs of nodes on their spatial relationships, resulting in different messages going to neighbours of the same node. To this end, we explore various ways of applying spatial conditioning under a multi-branch structure. Through extensive experimentation we demonstrate the advantages of spatial conditioning for the computation of the adjacency structure, messages and the refined graph features. In particular, we empirically show that as the quality of the bounding boxes increases, their coarse appearance features contribute relatively less to the disambiguation of interactions compared to the spatial information. Our method achieves an mAP of 31.33% on HICO-DET and 54.2% on V-COCO, significantly outperforming state-of-the-art on fine-tuned detections.

Dylan Campbell, Liu Liu, Stephen Gould

Blind Perspective-n-Point (PnP) is the problem of estimating the position and orientation of a camera relative to a scene, given 2D image points and 3D scene points, without prior knowledge of the 2D-3D correspondences. Solving for pose and correspondences simultaneously is extremely challenging since the search space is very large. Fortunately it is a coupled problem: the pose can be found easily given the correspondences and vice versa. Existing approaches assume that noisy correspondences are provided, that a good pose prior is available, or that the problem size is small. We instead propose the first fully end-to-end trainable network for solving the blind PnP problem efficiently and globally, that is, without the need for pose priors. We make use of recent results in differentiating optimization problems to incorporate geometric model fitting into an end-to-end learning framework, including Sinkhorn, RANSAC and PnP algorithms. Our proposed approach significantly outperforms other methods on synthetic and real data.

Yizhak Ben-Shabat, Xin Yu, Fatemeh Sadat Saleh et al.

The availability of a large labeled dataset is a key requirement for applying deep learning methods to solve various computer vision tasks. In the context of understanding human activities, existing public datasets, while large in size, are often limited to a single RGB camera and provide only per-frame or per-clip action annotations. To enable richer analysis and understanding of human activities, we introduce IKEA ASM -- a three million frame, multi-view, furniture assembly video dataset that includes depth, atomic actions, object segmentation, and human pose. Additionally, we benchmark prominent methods for video action recognition, object segmentation and human pose estimation tasks on this challenging dataset. The dataset enables the development of holistic methods, which integrate multi-modal and multi-view data to better perform on these tasks.

Yujiao Shi, Xin Yu, Dylan Campbell et al.

Cross-view geo-localization is the problem of estimating the position and orientation (latitude, longitude and azimuth angle) of a camera at ground level given a large-scale database of geo-tagged aerial (e.g., satellite) images. Existing approaches treat the task as a pure location estimation problem by learning discriminative feature descriptors, but neglect orientation alignment. It is well-recognized that knowing the orientation between ground and aerial images can significantly reduce matching ambiguity between these two views, especially when the ground-level images have a limited Field of View (FoV) instead of a full field-of-view panorama. Therefore, we design a Dynamic Similarity Matching network to estimate cross-view orientation alignment during localization. In particular, we address the cross-view domain gap by applying a polar transform to the aerial images to approximately align the images up to an unknown azimuth angle. Then, a two-stream convolutional network is used to learn deep features from the ground and polar-transformed aerial images. Finally, we obtain the orientation by computing the correlation between cross-view features, which also provides a more accurate measure of feature similarity, improving location recall. Experiments on standard datasets demonstrate that our method significantly improves state-of-the-art performance. Remarkably, we improve the top-1 location recall rate on the CVUSA dataset by a factor of 1.5x for panoramas with known orientation, by a factor of 3.3x for panoramas with unknown orientation, and by a factor of 6x for 180-degree FoV images with unknown orientation.

Rodrigo Santa Cruz, Anoop Cherian, Basura Fernando et al.

This paper presents a framework to recognize temporal compositions of atomic actions in videos. Specifically, we propose to express temporal compositions of actions as semantic regular expressions and derive an inference framework using probabilistic automata to recognize complex actions as satisfying these expressions on the input video features. Our approach is different from existing works that either predict long-range complex activities as unordered sets of atomic actions, or retrieve videos using natural language sentences. Instead, the proposed approach allows recognizing complex fine-grained activities using only pretrained action classifiers, without requiring any additional data, annotations or neural network training. To evaluate the potential of our approach, we provide experiments on synthetic datasets and challenging real action recognition datasets, such as MultiTHUMOS and Charades. We conclude that the proposed approach can extend state-of-the-art primitive action classifiers to vastly more complex activities without large performance degradation.

Liu Liu, Dylan Campbell, Hongdong Li et al.

Conventional absolute camera pose via a Perspective-n-Point (PnP) solver often assumes that the correspondences between 2D image pixels and 3D points are given. When the correspondences between 2D and 3D points are not known a priori, the task becomes the much more challenging blind PnP problem. This paper proposes a deep CNN model which simultaneously solves for both the 6-DoF absolute camera pose and 2D--3D correspondences. Our model comprises three neural modules connected in sequence. First, a two-stream PointNet-inspired network is applied directly to both the 2D image keypoints and the 3D scene points in order to extract discriminative point-wise features harnessing both local and contextual information. Second, a global feature matching module is employed to estimate a matchability matrix among all 2D--3D pairs. Third, the obtained matchability matrix is fed into a classification module to disambiguate inlier matches. The entire network is trained end-to-end, followed by a robust model fitting (P3P-RANSAC) at test time only to recover the 6-DoF camera pose. Extensive tests on both real and simulated data have shown that our method substantially outperforms existing approaches, and is capable of processing thousands of points a second with the state-of-the-art accuracy.

Shihao Jiang, Dylan Campbell, Miaomiao Liu et al.

We address the problem of joint optical flow and camera motion estimation in rigid scenes by incorporating geometric constraints into an unsupervised deep learning framework. Unlike existing approaches which rely on brightness constancy and local smoothness for optical flow estimation, we exploit the global relationship between optical flow and camera motion using epipolar geometry. In particular, we formulate the prediction of optical flow and camera motion as a bi-level optimization problem, consisting of an upper-level problem to estimate the flow that conforms to the predicted camera motion, and a lower-level problem to estimate the camera motion given the predicted optical flow. We use implicit differentiation to enable back-propagation through the lower-level geometric optimization layer independent of its implementation, allowing end-to-end training of the network. With globally-enforced geometric constraints, we are able to improve the quality of the estimated optical flow in challenging scenarios and obtain better camera motion estimates compared to other unsupervised learning methods.

Ali Cheraghian, Shafin Rahman, Dylan Campbell et al.

Zero-shot learning, the task of learning to recognize new classes not seen during training, has received considerable attention in the case of 2D image classification. However despite the increasing ubiquity of 3D sensors, the corresponding 3D point cloud classification problem has not been meaningfully explored and introduces new challenges. This paper extends, for the first time, transductive Zero-Shot Learning (ZSL) and Generalized Zero-Shot Learning (GZSL) approaches to the domain of 3D point cloud classification. To this end, a novel triplet loss is developed that takes advantage of unlabeled test data. While designed for the task of 3D point cloud classification, the method is also shown to be applicable to the more common use-case of 2D image classification. An extensive set of experiments is carried out, establishing state-of-the-art for ZSL and GZSL in the 3D point cloud domain, as well as demonstrating the applicability of the approach to the image domain.

Stephen Gould, Richard Hartley, Dylan Campbell

We explore a new class of end-to-end learnable models wherein data processing nodes (or network layers) are defined in terms of desired behavior rather than an explicit forward function. Specifically, the forward function is implicitly defined as the solution to a mathematical optimization problem. Consistent with nomenclature in the programming languages community, we name these models deep declarative networks. Importantly, we show that the class of deep declarative networks subsumes current deep learning models. Moreover, invoking the implicit function theorem, we show how gradients can be back-propagated through many declaratively defined data processing nodes thereby enabling end-to-end learning. We show how these declarative processing nodes can be implemented in the popular PyTorch deep learning software library allowing declarative and imperative nodes to co-exist within the same network. We also provide numerous insights and illustrative examples of declarative nodes and demonstrate their application for image and point cloud classification tasks.

Ali Cheraghian, Shafin Rahman, Dylan Campbell et al.

The development of advanced 3D sensors has enabled many objects to be captured in the wild at a large scale, and a 3D object recognition system may therefore encounter many objects for which the system has received no training. Zero-Shot Learning (ZSL) approaches can assist such systems in recognizing previously unseen objects. Applying ZSL to 3D point cloud objects is an emerging topic in the area of 3D vision, however, a significant problem that ZSL often suffers from is the so-called hubness problem, which is when a model is biased to predict only a few particular labels for most of the test instances. We observe that this hubness problem is even more severe for 3D recognition than for 2D recognition. One reason for this is that in 2D one can use pre-trained networks trained on large datasets like ImageNet, which produces high-quality features. However, in the 3D case there are no such large-scale, labelled datasets available for pre-training which means that the extracted 3D features are of poorer quality which, in turn, exacerbates the hubness problem. In this paper, we therefore propose a loss to specifically address the hubness problem. Our proposed method is effective for both Zero-Shot and Generalized Zero-Shot Learning, and we perform extensive evaluations on the challenging datasets ModelNet40, ModelNet10, McGill and SHREC2015. A new state-of-the-art result for both zero-shot tasks in the 3D case is established.

Dylan Campbell, Lars Petersson, Laurent Kneip et al.

Determining the position and orientation of a calibrated camera from a single image with respect to a 3D model is an essential task for many applications. When 2D-3D correspondences can be obtained reliably, perspective-n-point solvers can be used to recover the camera pose. However, without the pose it is non-trivial to find cross-modality correspondences between 2D images and 3D models, particularly when the latter only contains geometric information. Consequently, the problem becomes one of estimating pose and correspondences jointly. Since outliers and local optima are so prevalent, robust objective functions and global search strategies are desirable. Hence, we cast the problem as a 2D-3D mixture model alignment task and propose the first globally-optimal solution to this formulation under the robust $L_2$ distance between mixture distributions. We search the 6D camera pose space using branch-and-bound, which requires novel bounds, to obviate the need for a pose estimate and guarantee global optimality. To accelerate convergence, we integrate local optimization, implement GPU bound computations, and provide an intuitive way to incorporate side information such as semantic labels. The algorithm is evaluated on challenging synthetic and real datasets, outperforming existing approaches and reliably converging to the global optimum.

Dylan Campbell, Lars Petersson, Laurent Kneip et al.

Estimating the 6-DoF pose of a camera from a single image relative to a pre-computed 3D point-set is an important task for many computer vision applications. Perspective-n-Point (PnP) solvers are routinely used for camera pose estimation, provided that a good quality set of 2D-3D feature correspondences are known beforehand. However, finding optimal correspondences between 2D key-points and a 3D point-set is non-trivial, especially when only geometric (position) information is known. Existing approaches to the simultaneous pose and correspondence problem use local optimisation, and are therefore unlikely to find the optimal solution without a good pose initialisation, or introduce restrictive assumptions. Since a large proportion of outliers are common for this problem, we instead propose a globally-optimal inlier set cardinality maximisation approach which jointly estimates optimal camera pose and optimal correspondences. Our approach employs branch-and-bound to search the 6D space of camera poses, guaranteeing global optimality without requiring a pose prior. The geometry of SE(3) is used to find novel upper and lower bounds for the number of inliers and local optimisation is integrated to accelerate convergence. The evaluation empirically supports the optimality proof and shows that the method performs much more robustly than existing approaches, including on a large-scale outdoor data-set.

Jiaolong Yang, Hongdong Li, Dylan Campbell et al.

The Iterative Closest Point (ICP) algorithm is one of the most widely used methods for point-set registration. However, being based on local iterative optimization, ICP is known to be susceptible to local minima. Its performance critically relies on the quality of the initialization and only local optimality is guaranteed. This paper presents the first globally optimal algorithm, named Go-ICP, for Euclidean (rigid) registration of two 3D point-sets under the L2 error metric defined in ICP. The Go-ICP method is based on a branch-and-bound (BnB) scheme that searches the entire 3D motion space SE(3). By exploiting the special structure of SE(3) geometry, we derive novel upper and lower bounds for the registration error function. Local ICP is integrated into the BnB scheme, which speeds up the new method while guaranteeing global optimality. We also discuss extensions, addressing the issue of outlier robustness. The evaluation demonstrates that the proposed method is able to produce reliable registration results regardless of the initialization. Go-ICP can be applied in scenarios where an optimal solution is desirable or where a good initialization is not always available.

Dylan Campbell, Lars Petersson

Gaussian mixture alignment is a family of approaches that are frequently used for robustly solving the point-set registration problem. However, since they use local optimisation, they are susceptible to local minima and can only guarantee local optimality. Consequently, their accuracy is strongly dependent on the quality of the initialisation. This paper presents the first globally-optimal solution to the 3D rigid Gaussian mixture alignment problem under the L2 distance between mixtures. The algorithm, named GOGMA, employs a branch-and-bound approach to search the space of 3D rigid motions SE(3), guaranteeing global optimality regardless of the initialisation. The geometry of SE(3) was used to find novel upper and lower bounds for the objective function and local optimisation was integrated into the scheme to accelerate convergence without voiding the optimality guarantee. The evaluation empirically supported the optimality proof and showed that the method performed much more robustly on two challenging datasets than an existing globally-optimal registration solution.

Dylan Campbell, Lars Petersson

This paper presents a framework for rigid point-set registration and merging using a robust continuous data representation. Our point-set representation is constructed by training a one-class support vector machine with a Gaussian radial basis function kernel and subsequently approximating the output function with a Gaussian mixture model. We leverage the representation's sparse parametrisation and robustness to noise, outliers and occlusions in an efficient registration algorithm that minimises the L2 distance between our support vector--parametrised Gaussian mixtures. In contrast, existing techniques, such as Iterative Closest Point and Gaussian mixture approaches, manifest a narrower region of convergence and are less robust to occlusions and missing data, as demonstrated in the evaluation on a range of 2D and 3D datasets. Finally, we present a novel algorithm, GMMerge, that parsimoniously and equitably merges aligned mixture models, allowing the framework to be used for reconstruction and mapping.