labelCloud: A Lightweight Domain-Independent Labeling Tool for 3D Object Detection in Point Clouds

Sager, Christoph; Zschech, Patrick; Kühl, Niklas

doi:10.14733/cadconfp.2021.319-323

Cited by 8 publications

(4 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further, methods of active or reinforcement learning might, in the long run, reduce the amount of manual labelling. However, new media types, such as depth images or point cloud data, could lead to new challenges in labelling tasks (e.g., Sager et al, 2021). Economic considerations of labour costs (crowdsourcing) and opportunity costs (misclassification) could allow the calculation of an optimal cost-accuracy tradeoff.…”

Section: Discussionmentioning

confidence: 99%

A survey of image labelling for computer vision applications

Sager

Janiesch

Zschech

2021

Journal of Business Analytics

Self Cite

View full text Add to dashboard Cite

Supervised machine learning methods for image analysis require large amounts of labelled training data to solve computer vision problems. The recent rise of deep learning algorithms for recognising image content has led to the emergence of many ad-hoc labelling tools. With this survey, we capture and systematise the commonalities as well as the distinctions between existing image labelling software. We perform a structured literature review to compile the underlying concepts and features of image labelling software such as annotation expressiveness and degree of automation. We structure the manual labelling task by its organisation of work, user interface design options, and user support techniques to derive a systematisation schema for this survey. Applying it to available software and the body of literature, enabled us to uncover several application archetypes and key domains such as image retrieval or instance identification in healthcare or television.

show abstract

Section: Discussionmentioning

confidence: 99%

A survey of image labelling for computer vision applications

Sager

Janiesch

Zschech

2021

Journal of Business Analytics

Self Cite

View full text Add to dashboard Cite

show abstract

“…To train and evaluate pose estimation on AVD we first provide pose annotation for the main object categories of sofa, table, desk, bed, and chair. We first get the dense 3D dense point-cloud of each scene using each scene RGB and depth images and annotate the 3D bounding boxes for objects using LabelCloud tool [21], the example is in Figure 4c. We then project corners of 3D bounding-boxes in world coordinate are projected back to the image plane using the transformation matrix from world to camera, Figure 4d, base on the following equation: [X c ,Y c , Z c ] T is a point in the camera coordinate frame and X w = [X w ,Y w , Z w ] T is the point in the world coordinate frame.…”

Section: Active Vision Dataset Pose Labelingmentioning

confidence: 99%

Object Pose Estimation using Mid-level Visual Representations

Nejatishahidin¹,

Pooya²,

Košecká³

2022

Preprint

View full text Add to dashboard Cite

This work proposes a novel pose estimation model for object categories that can be effectively transferred to previously unseen environments. The deep convolutional network models (CNN) for pose estimation are typically trained and evaluated on datasets specifically curated for object detection, pose estimation, or 3D reconstruction, which requires large amounts of training data. In this work, we propose a model for pose estimation that can be trained with small amount of data and is built on the top of generic mid-level representations [33] (e.g. surface normal estimation and re-shading). These representations are trained on a large dataset without requiring pose and object annotations. Later on, the predictions are refined with a small CNN neural network that exploits object masks and silhouette retrieval. The presented approach achieves superior performance on the Pix3D dataset [26] and shows nearly 35% improvement over the existing models when only 25% of the training data is available. We show that the approach is favorable when it comes to generalization and transfer to novel environments. Towards this end, we introduce a new pose estimation benchmark for commonly encountered furniture categories on challenging Active Vision Dataset [1] and evaluated the models trained on the Pix3D dataset.

show abstract

“…To obtain the 3D bounding box, region growing on the projected depth points of 2D image detection followed by box estimation was used to create ground truth automatically. This step was followed by a 3D orientation and manual error correction of pedestrians using LabelCloud (Sager et al, 2021). A total of 855 pedestrian instances (3D bounding boxes) were annotated using the above mentioned procedure with the pedestrian orientation distribution shown in Figure 4 for the complete dataset.…”

Section: Annotationmentioning

confidence: 99%

Improving 3d Pedestrian Detection for Wearable Sensor Data With 2d Human Pose

Kamalasanan

Sester

2022

ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. Collisions and safety are important concepts when dealing with urban designs like shared spaces. As pedestrians (especially the elderly and disabled people) are more vulnerable to accidents, realising an intelligent mobility aid to avoid collisions is a direction of research that could improve safety using a wearable device. Also, with the improvements in technologies for visualisation and their capabilities to render 3D virtual content, AR devices could be used to realise virtual infrastructure and virtual traffic systems. Such devices (e.g., Hololens) scan the environment using stereo and ToF (Time-of-Flight) sensors, which in principle can be used to detect surrounding objects, including dynamic agents such as pedestrians. This can be used as basis to predict collisions. To envision an AR device as a safety aid and demonstrate its 3D object detection capability (in particular: pedestrian detection), we propose an improvement to the 3D object detection framework Frustum Pointnet with human pose and apply it on the data from an AR device. Using the data from such a device in an indoor setting, we conducted a comparative study to investigate how high level 2D human pose features in our approach could help to improve the detection performance of orientated 3D pedestrian instances over Frustum Pointnet.

show abstract

labelCloud: A Lightweight Domain-Independent Labeling Tool for 3D Object Detection in Point Clouds

Cited by 8 publications

References 4 publications

A survey of image labelling for computer vision applications

A survey of image labelling for computer vision applications

Object Pose Estimation using Mid-level Visual Representations

Improving 3d Pedestrian Detection for Wearable Sensor Data With 2d Human Pose

Contact Info

Product

Resources

About