Fast algorithms for large scale conditional 3D prediction

Bo, Liefeng; Sminchisescu, Cristian; Kanaujia, Atul; Metaxas, Dimitris N.

doi:10.1109/cvpr.2008.4587578

Cited by 70 publications

(70 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In recent years, a tendency towards learning based methods has been observed, to overcome the computational burden of searching the high-dimensional human pose space. While some systems try to infer a direct mapping from observed image features to articulated human poses [9,6], others learn priors for human dynamics to provide a better prediction [19]. Both of these directions are unable to detect arbitrary, previously unobserved motions, making it difficult to apply them to the observation of everyday activities.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

The TUM Kitchen Data Set of everyday manipulation activities for motion tracking and action recognition

Tenorth

Bandouch

Beetz

2009

2009 IEEE 12th International Conference on Computer Vision Workshops, ICCV Workshops

168

162

View full text Add to dashboard Cite

We introduce the publicly available TUM Kitchen Data Set as a comprehensive collection of activity sequences recorded in a kitchen environment equipped with multiple complementary sensors. The recorded data consists of observations of naturally performed manipulation tasks as encountered in everyday activities of human life. Several instances of a table-setting task were performed by different subjects, involving the manipulation of objects and the environment. We provide the original video sequences, fullbody motion capture data recorded by a markerless motion tracker, RFID tag readings and magnetic sensor readings from objects and the environment, as well as corresponding action labels. In this paper, we both describe how the data was computed, in particular the motion tracker and the labeling, and give examples what it can be used for. We present first results of an automatic method for segmenting the observed motions into semantic classes, and describe how the data can be integrated in a knowledge-based framework for reasoning about the observations.

show abstract

Section: Related Workmentioning

confidence: 99%

“…Data is provided in the BVH file format 6 , which contains the 6 DOF pose and the joint angle values as defined in Fig. 2.…”

Section: Datamentioning

confidence: 99%

The TUM Kitchen Data Set of everyday manipulation activities for motion tracking and action recognition

Tenorth

Bandouch

Beetz

2009

2009 IEEE 12th International Conference on Computer Vision Workshops, ICCV Workshops

168

162

View full text Add to dashboard Cite

show abstract

“…[20]. Recently, many novel MOE methods are proposed to handle high dimensional data [21][22][23]. In this paper, a new trace norm regularized MOE model is proposed.…”

Section: Introductionmentioning

confidence: 99%

Multi-task feature learning by using trace norm regularization

Zhang

et al. 2017

Open Physics

View full text Add to dashboard Cite

Abstract:Multi-task learning can extract the correlation of multiple related machine learning problems to improve performance. This paper considers applying the multi-task learning method to learn a single task. We propose a new learning approach, which employs the mixture of expert model to divide a learning task into several related subtasks, and then uses the trace norm regularization to extract common feature representation of these sub-tasks. A nonlinear extension of this approach by using kernel is also provided. Experiments conducted on both simulated and real data sets demonstrate the advantage of the proposed approach.

show abstract

“…For a given test image, a similarity search is performed to find similar candidates in training set and then obtain estimated poses by interpolating from their poses [2]. On the other hand, learning-based approaches learn the direct mapping from image observations to pose space using training samples [3][4][5]. While generative methods can infer poses with better precision than discriminative ones, discriminative approaches have the advantage in execution time.…”

Section: Introductionmentioning

confidence: 99%

“…Furthermore, it is computationally expensive to compute these functions and thus makes them unsuitable for inferring the hidden poses. In contrast, discriminative methods do not assume a particular human body model, and they can be further categorized as example-based [2] and learning-based [3][4][5]. Example-based approaches store a set of training samples along with their corresponding pose descriptors.…”

Section: Introductionmentioning

confidence: 99%

Estimating Human Pose from Occluded Images

Huang

Yang

2010

Computer Vision – ACCV 2009

View full text Add to dashboard Cite

Abstract. We address the problem of recovering 3D human pose from single 2D images, in which the pose estimation problem is formulated as a direct nonlinear regression from image observation to 3D joint positions. One key issue that has not been addressed in the literature is how to estimate 3D pose when humans in the scenes are partially or heavily occluded. When occlusions occur, features extracted from image observations (e.g., silhouettes-based shape features, histogram of oriented gradient, etc.) are seriously corrupted, and consequently the regressor (trained on un-occluded images) is unable to estimate pose states correctly. In this paper, we present a method that is capable of handling occlusions using sparse signal representations, in which each test sample is represented as a compact linear combination of training samples. The sparsest solution can then be efficiently obtained by solving a convex optimization problem with certain norms (such as l1-norm). The corrupted test image can be recovered with a sparse linear combination of un-occluded training images which can then be used for estimating human pose correctly (as if no occlusions exist). We also show that the proposed approach implicitly performs relevant feature selection with un-occluded test images. Experimental results on synthetic and real data sets bear out our theory that with sparse representation 3D human pose can be robustly estimated when humans are partially or heavily occluded in the scenes.

show abstract

Fast algorithms for large scale conditional 3D prediction

Abstract: The potential success of discriminative learning approaches

Cited by 70 publications

References 21 publications

The TUM Kitchen Data Set of everyday manipulation activities for motion tracking and action recognition

The TUM Kitchen Data Set of everyday manipulation activities for motion tracking and action recognition

Multi-task feature learning by using trace norm regularization

Estimating Human Pose from Occluded Images

Contact Info

Product

Resources

About