Supervised deep learning methods have shown promising results for the task of monocular depth estimation; but acquiring ground truth is costly, and prone to noise as well inaccuracies. While synthetic datasets been used circumvent above problems, resultant models do not generalize natural scenes due inherent domain shift. Recent adversarial approaches adaption performed in mitigating differences between source target domains. But these are mostly limited a classification setup scale...

There is a strong incentive to develop versatile learning techniques that can transfer the knowledge of class-separability from labeled source domain an unlabeled target in presence domain-shift. Existing adaptation (DA) approaches are not equipped for practical DA scenarios as result their reliance on source-target label-set relationship (e.g. Closed-set, Open-set or Partial DA). Furthermore, almost all prior unsupervised works require coexistence and samples even during deployment, making...

Human motion prediction model has applications in various fields of computer vision. Without taking into account the inherent stochasticity future pose dynamics, such methods often converges to a deterministic undesired mean multiple probable outcomes. Devoid this, we propose novel probabilistic generative approach called Bidirectional GAN, or BiHMP-GAN. To be able generate human-pose sequences, conditioned on given starting sequence, introduce random extrinsic factor r, drawn from...

Unsupervised domain adaptation (DA) has gained substantial interest in semantic segmentation. However, almost all prior arts assume concurrent access to both labeled source and unlabeled target, making them unsuitable for scenarios demanding source-free adaptation. In this work <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> , we enable DA by partitioning the task into two: a) source-only generalization b) target Towards former, provide...

Camera captured human pose is an outcome of several sources variation. Performance supervised 3D estimation approaches comes at the cost dispensing with variations, such as shape and appearance, that may be useful for solving other related tasks. As a result, learned model not only inculcates task-bias but also dataset-bias because its strong reliance on annotated samples, which holds true weakly-supervised models. Acknowledging this, we propose self-supervised learning framework to...

There has been a tremendous progress in Domain Adaptation (DA) for visual recognition tasks. Particularly, open-set DA gained considerable attention wherein the target domain contains additional unseen categories. Existing approaches demand access to labeled source dataset along with unlabeled instances. However, this reliance on co-existing and data is highly impractical scenarios where data-sharing restricted due its proprietary nature or privacy concerns. Addressing this, we introduce...

An unsupervised human action modeling framework can provide useful pose-sequence representation, which be utilized in a variety of pose analysis applications. In this work we propose novel temporal framework, embed the dynamics 3D human-skeleton joints to latent space an efficient manner. contrast end-to-end explored by previous works, disentangle task individual representation learning from actions as sequence embeddings. order realize continuous embedding manifold along with better...

The advances in monocular 3D human pose estimation are dominated by supervised techniques that require large-scale 2D/3D annotations. Such methods often behave erratically the absence of any provision to discard unfamiliar out-of-distribution data. To this end, we cast learning as an unsupervised domain adaptation problem. We introduce MRP-Net <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> Project page:...

Aiming towards human-level generalization, there is a need to explore adaptable representation learning methods with greater transferability. Most existing approaches independently address task-transferability and cross-domain adaptation, resulting in limited generalization. In this paper, we propose UM-Adapt - unified framework effectively perform unsupervised domain adaptation for spatially-structured prediction tasks, simultaneously maintaining balanced performance across individual tasks...

Freehand sketching is an inherently sequential process. Yet, most approaches for hand-drawn sketch recognition either ignore this aspect or exploit it in ad-hoc manner. In our work, we propose a recurrent neural network architecture object which exploits the long-term and structural regularities stroke data scalable Specifically, introduce Gated Recurrent Unit based framework leverages deep features weighted per-timestep loss to achieve state-of-the-art results on large database of freehand...

Estimation of 3D human pose from monocular image has gained considerable attention, as a key step to several human-centric applications. However, generalizability estimation models developed using supervision on large-scale in-studio datasets remains questionable, these often perform unsatisfactorily unseen in-the-wild environments. Though weakly-supervised have been proposed address this shortcoming, performance such relies availability paired some related task, 2D or multi-view pairs. In...

Multi-Source Domain Adaptation (MSDA) deals with the transfer of task knowledge from multiple labeled source domains to an unlabeled target domain, under a domain-shift. Existing methods aim minimize this domain-shift using auxiliary distribution alignment objectives. In work, we present different perspective MSDA wherein deep models are observed implicitly align label supervision. Thus, utilize implicit without additional training objectives perform adaptation. To end, use pseudo-labeled...

Conventional domain adaptation (DA) techniques aim to improve transferability by learning domain-invariant representations; while concurrently preserving the task-discriminability knowledge gathered from labeled source data. However, requirement of simultaneous access and unlabeled target renders them unsuitable for challenging source-free DA setting. The trivial solution realizing an effective original generic mapping improves but degrades task discriminability. Upon analyzing hurdles both...

Modeling dynamics of human motion is one the most challenging sequence modeling problem, with diverse applications in animation industry, human-robot interaction, motion-based surveillance, etc. Available attempts to use auto-regressive techniques for long-term single-person generation usually fails, resulting stagnated or divergence unrealistic pose patterns. In this paper, we propose a novel cross-conditioned recurrent framework targeting synthesis inter-person interactions beyond several...

Conventional Domain Adaptation (DA) methods aim to learn domain-invariant feature representations improve the target adaptation performance. However, we motivate that domain-specificity is equally important since in-domain trained models hold crucial domain-specific properties are beneficial for adaptation. Hence, propose build a framework supports disentanglement and learning of factors task-specific in unified model. Motivated by success vision transformers several multi-modal problems,...

We present a biologically inspired recurrent neural network (RNN) that efficiently detects changes in natural images. The model features sparse, topographic connectivity (st-RNN), closely modeled on the circuit architecture of “midbrain attention network.” deployed st-RNN challenging change blindness task, which must be detected discontinuous sequence Compared with conventional RNN, learned 9x faster and achieved state-of-the-art performance 15x fewer connections. An analysis low-dimensional...

Despite the remarkable success of generative adversarial networks, their performance seems less impressive for diverse training sets, requiring learning discontinuous mapping functions. Though multi-mode prior or multi-generator models have been proposed to alleviate this problem, such approaches may fail depending on empirically chosen initial mode components. In contrast bottom-up approaches, we present GAN-Tree, which follows a hierarchical divisive strategy address multi-modal data....

Universal Domain Adaptation (UniDA) deals with the problem of knowledge transfer between two datasets domain-shift as well category-shift. The goal is to categorize unlabeled target samples, either into one "known" categories or a single "unknown" category. A major in UniDA negative transfer, i.e. misalignment and classes. To this end, we first uncover an intriguing tradeoff negative-transfer-risk domain-invariance exhibited at different layers deep network. It turns out can strike balance...