Beat the MTurkers: Automatic Image Labeling from Weak 3D Supervision

By Liang-Chieh Chen, Sanja Fidler, Raquel Urtasun. 2014

87 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computational Mechanics, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

Iteratively Trained Interactive Segmentation (2018, methodology): “For the experiments on KITTI [12], we use 741 cars annotated at the pixel level provided by [6].”
Weakly-and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation (2015, background): “[4] attained human-level accuracy for car segmentation by using 3D bounding boxes.”
Monocular Object Instance Segmentation and Depth Ordering with CNNs (2015, neutral): “Our experimental evaluations show that our network is capable of accurate prediction for up to 5 object instances in an image patch.”
Pixel-Level Encoding and Depth Layering for Instance-Level Semantic Labeling (2016, methodology): “We evaluated our approach on the KITTI object detection dataset [10] extended with instance-level segmentations [3,35] as well as Cityscapes [6].”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “CAD models to improve solutions for pose and viewpoint estimation [161, 162, 163, 164], object and object part detection [165, 166, 167, 168, 169, 170, 171, 172], segmentation and scene understanding [173, 174, 175]. The recent progresses in computer graphics and modern high-level generic graphics p”
Instance-Level Segmentation for Autonomous Driving with Deep Densely Connected MRFs (2015, methodology): “131 images from either our validation or test set have been manually annotated with pixel-wise instance labeling for Cars by [5].”
Box2Pix: Single-Shot Instance Segmentation by Assigning Pixels to Object Boxes (2018, methodology): “In the following, we evaluate our approach on two challenging datasets, namely Cityscapes [1] and KITTI [2], [5], [6] using suggested instance-aware metrics.”
Deep MANTA: A Coarse-to-Fine Many-Task Network for Joint 2D and 3D Vehicle Analysis from Monocular Image (2017, methodology): “We use the 3D CAD dataset provided by [11, 6] composed of M = 103 3D vehicle models for semi-automatic annotation.”
Semantic Instance Annotation of Street Scenes by 3D to 2D Label Transfer (2015, background): “Dynamic objects could be handled via 3D models [9,28] but as our dataset comprises little dynamic objects we leave this extension for future work.”
Object Instance Annotation With Deep Extreme Level Set Evolution (2019, methodology): “Early work on interactive annotation incorporated user strokes on both the foreground and background and used graph cut optimization to find objects [36, 6, 12].”
A General Pipeline for 3D Detection of Vehicles (2018, methodology): “We verified this and generalised the car models by normalising the cars in the 3D CAD dataset used in [7], [29] and [32].”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, methodology): “Towards this goal, Chen et al. (2014) use annotations from KITTI (Geiger et al. (2013)) and leverage 3D car CAD models to infer separate figure-ground segmentations for all cars in the image.”
Instance-Level Segmentation with Deep Densely Connected MRFs (2015, methodology): “R-CNN [9] was used in [10] to first generate object proposals, and then predictions from two CNNs (box and region-based) were fused to segment the object inside each box.”
Efficient Interactive Annotation of Segmentation Datasets with Polygon-RNN++ (2018, methodology): “It achieves human level agreement of 85% as reported by [5] by requiring only 2 clicks on average by the (simulated) annotator.”
BigDatasetGAN: Synthesizing ImageNet with Pixel-wise Annotations (2022, background): “Reducing labeling costs can be achieved in a variety of ways, such as interactive humanin-the-loop annotation [2, 50, 53, 64, 73], nearest-neighbor mask transfer [26, 51, 63, 70], or using weak supervision in the form of cheaper labels such as object boxes [7, 43, 45], scribbles [48, 69] or very coa”
Joint Graph Decomposition & Node Labeling: Problem, Algorithms, Applications (2016, methodology): “As there are different metrics used by related approaches, we report performance on the Cityscapes [4] and KITTI [2, 5] dataset using both proposed metrics.”
Learning to segment under various forms of weak supervision (2015, background): “3D bounding boxes as a form of weak supervision have been shown to produce human-level segmentation results [5].”
Weakly- and Semi-Supervised Learning of a DCNN for Semantic Image Segmentation (2015, methodology): “upervised setting. Bounding box annotations have been utilized for semantic segmentation by [39,43], while [15,21,41] describe schemes exploiting both image-level labels and bounding box annotations. [4] attained human-level accuracy for car segmentation by using 3D bounding boxes. Bounding box anno”