With the availability of egocentric 3D hand-object interaction datasets, there is increasing interest in developing unified models for hand-object pose estimation and action recognition. However, existing methods still struggle to recognise seen actions on unseen objects due to the limitations in representing object shape and movement using 3D bounding boxes. Additionally, the reliance on object templates at test time limits their generalisability to unseen objects. To address these challenges, we propose to leverage superquadrics as an alternative 3D object representation to bounding boxes and demonstrate their effectiveness on both template-free object reconstruction and action recognition tasks. Moreover, as we find that pure appearance-based methods can outperform the unified methods, the potential benefits from 3D geometric information remain unclear. Therefore, we study the compositionality of actions by considering a more challenging task where the training combinations of verbs and nouns do not overlap with the testing split. We extend H2O and FPHA datasets with compositional splits and design a novel collaborative learning framework that can explicitly reason about the geometric relations between hands and the manipulated object. Through extensive quantitative and qualitative evaluations, we demonstrate significant improvements over the state-of-the-arts in (compositional) action recognition.

Quanta image sensors, such as SPAD arrays, are an emerging sensor technology, producing 1-bit arrays representing photon detection events over exposures as short as a few nanoseconds. In practice, raw data are post-processed using heavy spatiotemporal binning to create more useful and interpretable images at the cost of degrading spatiotemporal resolution. In this work, we propose bit2bit, a new method for reconstructing high-quality image stacks at the original spatiotemporal resolution from sparse binary quanta image data. Inspired by recent work on Poisson denoising, we developed an algorithm that creates a dense image sequence from sparse binary photon data by predicting the photon arrival location probability distribution. However, due to the binary nature of the data, we show that the assumption of a Poisson distribution is inadequate. Instead, we model the process with a Bernoulli lattice process from the truncated Poisson. This leads to the proposal of a novel self-supervised solution based on a masked loss function. We evaluate our method using both simulated and real data. On simulated data from a conventional video, we achieve 34.35 mean PSNR with extremely photon-sparse binary input (<0.06 photons per pixel per frame). We also present a novel dataset containing a wide range of real SPAD high-speed videos under various challenging imaging conditions. The scenes cover strong/weak ambient light, strong motion, ultra-fast events, etc., which will be made available to the community, on which we demonstrate the promise of our approach. Both reconstruction quality and throughput substantially surpass the state-of-the-art methods (e.g., Quanta Burst Photography (QBP)). Our approach significantly enhances the visualization and usability of the data, enabling the application of existing analysis techniques.

Indirect Time-of-Flight (I-ToF) imaging is a widespread way of depth estimation for mobile devices due to its small size and affordable price. Previous works have mainly focused on quality improvement for I-ToF imaging especially curing the effect of Multi Path Interference (MPI). These investigations are typically done in specifically constrained scenarios at close distance, indoors and under little ambient light. Surprisingly little work has investigated I-ToF quality improvement in real-life scenarios where strong ambient light and far distances pose difficulties due to an extreme amount of induced shot noise and signal sparsity, caused by the attenuation with limited sensor power and light scattering. In this work, we propose a new learning based end-to-end depth prediction network which takes noisy raw I-ToF signals as well as an RGB image and fuses their latent representation based on a multi step approach involving both implicit and explicit alignment to predict a high quality long range depth map aligned to the RGB viewpoint. We test our approach on challenging real-world scenes and show more than 40% RMSE improvement on the final depth map compared to the baseline approach.

We study the problem of labelling effort for semantic segmentation of large-scale 3D point clouds. Existing works usually rely on densely annotated point-level semantic labels to provide supervision for network training. However, in real-world scenarios that contain billions of points, it is impractical and extremely costly to manually annotate every single point. In this paper, we first investigate whether dense 3D labels are truly required for learning meaningful semantic representations. Interestingly, we find that the segmentation performance of existing works only drops slightly given as few as 1% of the annotations. However, beyond this point (e.g. 1 per thousand and below) existing techniques fail catastrophically. To this end, we propose a new weak supervision method to implicitly augment the total amount of available supervision signals, by leveraging the semantic similarity between neighboring points. Extensive experiments demonstrate that the proposed Semantic Query Network (SQN) achieves state-of-the-art performance on six large-scale open datasets under weak supervision schemes, while requiring only 1000x fewer labeled points for training. The code is available at https://github.com/QingyongHu/SQN.

Artificial neural networks thrive in solving the classification problem for a particular rigid task, where the network resembles a static entity of knowledge, acquired through generalized learning behaviour from a distinct training phase. However, endeavours to extend this knowledge without targeting the original task usually result in a catastrophic forgetting of this task. Continual learning shifts this paradigm towards a network that can continually accumulate knowledge over different tasks without the need for retraining from scratch, with methods in particular aiming to alleviate forgetting. We focus on task-incremental classification, where tasks arrive in a batch-like fashion, and are delineated by clear boundaries. Our main contributions concern 1) a taxonomy and extensive overview of the state-of-the-art, 2) a novel framework to continually determine stability-plasticity trade-off of the continual learner, 3) a comprehensive experimental comparison of 10 state-of-the-art continual learning methods and 4 baselines. We empirically scrutinize which method performs best, both on balanced Tiny Imagenet and a large-scale unbalanced iNaturalist datasets. We study the influence of model capacity, weight decay and dropout regularization, and the order in which the tasks are presented, and qualitatively compare methods in terms of required memory, computation time and storage.

Human scene understanding uses a variety of visual and non-visual cues to perform inference on object types, poses, and relations. Physics is a rich and universal cue which we exploit to enhance scene understanding. We integrate the physical cue of stability into the learning process using a REINFORCE approach coupled to a physics engine, and apply this to the problem of producing the 3D bounding boxes and poses of objects in a scene. We first show that applying physics supervision to an existing scene understanding model increases performance, produces more stable predictions, and allows training to an equivalent performance level with fewer annotated training examples. We then present a novel architecture for 3D scene parsing named Prim R-CNN, learning to predict bounding boxes as well as their 3D size, translation, and rotation. With physics supervision, Prim R-CNN outperforms existing scene understanding approaches on this problem. Finally, we show that applying physics supervision on unlabeled real images improves real domain transfer of models training on synthetic data.

Multi-class object learning and detection is a challenging problem due to the large number of object classes and their high visual variability. Specialized detectors usuallyexcel in performance, while joint representations optimize sharing and reduce inference time -- but are complex to train. Conveniently, sequential class learning cuts down training time by transferring existing knowledge to novel classes, but cannot fully exploit the shareability of features among object classes and might depend on ordering of classes during learning. In hierarchical frameworks theseissues have been little explored. In this paper, we provide a rigorous experimental analysis of various multiple object class learning strategies within a generative hierarchical framework. Specifically, we propose, evaluate and compare threeimportant types of multi-class learning: 1.) independent training of individual categories, 2.) joint training of classes, and 3.) sequential learning of classes. We explore and compare their computational behavior (space and time) and detection performance as a function of the number of learned object classes on several recognition datasets. We show that sequential training achieves the best tradeoff between inference and training times at a comparable detection performance andcould thus be used to learn the classes on a larger scale.