For each training sequence hI0,...,Ini, Only-Occlusion randomly selects four indexes 0 < k1 k2 < k3 k4 < n to construct two disjoint frame subsequences hIk1,...,Ik2i and hIk3,...,Ik4i. Learning to merely compare detection features across consecutive frames would yield low accuracy since features in occluded frames are not observed. This strategy yields high consistency because it is unaffected by occluded intermediate frames. We select two indexes 0 < k5,k6 < n. Then, we randomly pick k5 and k6 such that k3 k5 k4 and k1 k6 k2, i.e., the hand-off for one tracker occurs when the other tracker observes a simulated occlusion.

Supervised training of deep neural networks on pairs of clean image and noisy measurement achieves state-of-the-art performance for many image reconstruction tasks, but such training pairs are difficult to collect. Self-supervised methods enable training based on noisy measurements only, without clean images. In this work, we investigate the cost of self-supervised training in terms of sample complexity for a class of self-supervised methods that enable the computation of unbiased estimates of gradients of the supervised loss, including noise2noise methods. We analytically show that a model trained with such self-supervised training is as good as the same model trained in a supervised fashion, but self-supervised training requires more examples than supervised training. We then study self-supervised denoising and accelerated MRI empirically and characterize the cost of self-supervised training in terms of the number of additional samples required, and find that the performance gap between self-supervised and supervised training vanishes as a function of the training examples, at a problem-dependent rate, as predicted by our theory.

We present a neural analysis and synthesis (NANSY) framework that can manipulate the voice, pitch, and speed of an arbitrary speech signal. Most of the previous works have focused on using information bottleneck to disentangle analysis features for controllable synthesis, which usually results in poor reconstruction quality. We address this issue by proposing a novel training strategy based on information perturbation. The idea is to perturb information in the original input signal (e.g., formant, pitch, and frequency response), thereby letting synthesis networks selectively take essential attributes to reconstruct the input signal. Because NANSY does not need any bottleneck structures, it enjoys both high reconstruction quality and controllability. Furthermore, NANSY does not require any labels associated with speech data such as text and speaker information, but rather uses a new set of analysis features, i.e., wav2vec feature and newly proposed pitch feature, Yingram, which allows for fully self-supervised training. Taking advantage of fully self-supervised training, NANSY can be easily extended to a multilingual setting by simply training it with a multilingual dataset. The experiments show that NANSY can achieve significant improvement in performance in several applications such as zero-shot voice conversion, pitch shift, and time-scale modification.

Supervised training of deep neural networks on pairs of clean image and noisy measurement achieves state-of-the-art performance for many image reconstruction tasks, but such training pairs are difficult to collect. Self-supervised methods enable training based on noisy measurements only, without clean images. In this work, we investigate the cost of self-supervised training in terms of sample complexity for a class of self-supervised methods that enable the computation of unbiased estimates of gradients of the supervised loss, including noise2noise methods. We analytically show that a model trained with such self-supervised training is as good as the same model trained in a supervised fashion, but self-supervised training requires more examples than supervised training. We then study self-supervised denoising and accelerated MRI empirically and characterize the cost of self-supervised training in terms of the number of additional samples required, and find that the performance gap between self-supervised and supervised training vanishes as a function of the training examples, at a problem-dependent rate, as predicted by our theory.

We present a neural analysis and synthesis (NANSY) framework that can manipulate the voice, pitch, and speed of an arbitrary speech signal. Most of the previous works have focused on using information bottleneck to disentangle analysis features for controllable synthesis, which usually results in poor reconstruction quality. We address this issue by proposing a novel training strategy based on information perturbation. The idea is to perturb information in the original input signal (e.g., formant, pitch, and frequency response), thereby letting synthesis networks selectively take essential attributes to reconstruct the input signal. Because NANSY does not need any bottleneck structures, it enjoys both high reconstruction quality and controllability.

Significant uncertainty in climate prediction and cloud physics is tied to observational gaps relating to shallow scattered clouds. Addressing these challenges requires remote sensing of their three-dimensional (3D) heterogeneous volumetric scattering content. This calls for passive scattering computed tomography (CT). We design a learning-based model (ProbCT) to achieve CT of such clouds, based on noisy multi-view spaceborne images. ProbCT infers - for the first time - the posterior probability distribution of the heterogeneous extinction coefficient, per 3D location. This yields arbitrary valuable statistics, e.g., the 3D field of the most probable extinction and its uncertainty. ProbCT uses a neural-field representation, making essentially real-time inference. ProbCT undergoes supervised training by a new labeled multi-class database of physics-based volumetric fields of clouds and their corresponding images. To improve out-of-distribution inference, we incorporate self-supervised learning through differential rendering. We demonstrate the approach in simulations and on real-world data, and indicate the relevance of 3D recovery and uncertainty to precipitation and renewable energy.

We consider a certifiable object pose estimation problem, where -- given a partial point cloud of an object -- the goal is to not only estimate the object pose, but also to provide a certificate of correctness for the resulting estimate. Our first contribution is a general theory of certification for end-to-end perception models. In particular, we introduce the notion of $\zeta$-correctness, which bounds the distance between an estimate and the ground truth. We show that $\zeta$-correctness can be assessed by implementing two certificates: (i) a certificate of observable correctness, that asserts if the model output is consistent with the input data and prior information, (ii) a certificate of non-degeneracy, that asserts whether the input data is sufficient to compute a unique estimate. Our second contribution is to apply this theory and design a new learning-based certifiable pose estimator. We propose C-3PO, a semantic-keypoint-based pose estimation model, augmented with the two certificates, to solve the certifiable pose estimation problem. C-3PO also includes a keypoint corrector, implemented as a differentiable optimization layer, that can correct large detection errors (e.g. due to the sim-to-real gap). Our third contribution is a novel self-supervised training approach that uses our certificate of observable correctness to provide the supervisory signal to C-3PO during training. In it, the model trains only on the observably correct input-output pairs, in each training iteration. As training progresses, we see that the observably correct input-output pairs grow, eventually reaching near 100% in many cases. Our experiments show that (i) standard semantic-keypoint-based methods outperform more recent alternatives, (ii) C-3PO further improves performance and significantly outperforms all the baselines, and (iii) C-3PO's certificates are able to discern correct pose estimates.