14 Hits in 3.5 sec

MSeg: A Composite Dataset for Multi-Domain Semantic Segmentation

John Lambert, Zhuang Liu, Ozan Sener, James Hays, Vladlen Koltun
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
Input image Ground truth ADE20K model Mapillary model COCO model MSeg model Figure 1: MSeg unifies multiple semantic segmentation datasets by reconciling their taxonomies and resolving incompatible annotations  ...  This enables training models that perform consistently across domains and generalize better.  ...  Conclusion We presented a composite dataset for multi-domain semantic segmentation. To construct the composite dataset, we reconciled the taxonomies of seven semantic segmentation datasets.  ... 
doi:10.1109/cvpr42600.2020.00295 dblp:conf/cvpr/LambertLSHK20 fatcat:nxepr7qrwbgqzhcsuiqlp7c7oi

Diffuser: Multi-View 2D-to-3D Label Diffusion for Semantic Scene Segmentation

Ruben Mascaro, Lucas Teixeira, Margarita Chli
2021 2021 IEEE International Conference on Robotics and Automation (ICRA)  
Despite recent advances in deep learning, its application to multi-domain 3D semantic segmentation typically suffers from the lack of extensive enough annotated 3D datasets.  ...  In this paper, we present 'Diffuser', a novel and efficient multi-view fusion framework that leverages 2D semantic segmentation of multiple image views of a scene to produce a consistent and refined 3D  ...  For both the indoor and outdoor datasets, we use a network based on the HRNet-W48 [30] architecture and trained on MSeg [2] , a composite, multi-domain semantic segmentation dataset designed for training  ... 
doi:10.1109/icra48506.2021.9561801 fatcat:pzgokupxyvfi3lwtedygxthypq

Automatic universal taxonomies for multi-domain semantic segmentation [article]

Petra Bevandić, Siniša Šegvić
2022 arXiv   pre-print
Training semantic segmentation models on multiple datasets has sparked a lot of recent interest in the computer vision community.  ...  This interest has been motivated by expensive annotations and a desire to achieve proficiency across multiple visual domains.  ...  We have successfully applied this procedure to the MSeg dataset collection. Experiments We train semantic segmentation models in multi-domain setups.  ... 
arXiv:2207.08445v1 fatcat:khoy4qro7zb57fo76aonmjojme

Dense outlier detection and open-set recognition based on training with noisy negative images [article]

Petra Bevandić, Ivan Krešo, Marin Oršić, Siniša Šegvić
2022 arXiv   pre-print
Second, we assume that there exists a general-purpose dataset which is much more diverse than the inlier dataset (e.g.~ImageNet-1k).  ...  Deep convolutional models often produce inadequate predictions for inputs foreign to the training distribution.  ...  The LDN OE model has a single C-way multi-class head and uses max-softmax for outlier detection. The LDN BIN and LDN BIN JS models have separate heads for semantic segmentation and outlier detection.  ... 
arXiv:2101.09193v2 fatcat:zux3txayvzdwpmbmw4rdytnrf4

Semantic Segmentation on Multiple Visual Domains [article]

Floris Naber
2021 arXiv   pre-print
Semantic segmentation models only perform well on the domain they are trained on and datasets for training are scarce and often have a small label-spaces, because the pixel level annotations required are  ...  In this paper a method for this is proposed for the datasets Cityscapes, SUIM and SUN RGB-D, by creating a label-space that spans all classes of the datasets.  ...  Multi-domain semantic segmentation is an issue that still has no good semi-automated solution.  ... 
arXiv:2107.04326v1 fatcat:t7zwcdae7ze2zfcqoukuabty5i

Exploring Event-driven Dynamic Context for Accident Scene Segmentation [article]

Jiaming Zhang, Kailun Yang, Rainer Stiefelhagen
2021 arXiv   pre-print
The robustness of semantic segmentation on edge cases of traffic scene is a vital factor for the safety of intelligent transportation.  ...  Our approach achieves +8.2% performance gain on the proposed accident dataset, exceeding more than 20 state-of-the-art semantic segmentation methods.  ...  MSeg [47] constructed a composite dataset with the aim of facilitating generalization for multi-domain semantic segmentation.  ... 
arXiv:2112.05006v1 fatcat:xyjr5tiysjgdxbvx2rk2huwdj4

BigDetection: A Large-scale Benchmark for Improved Object Detector Pre-training [article]

Likun Cai, Zhi Zhang, Yi Zhu, Li Zhang, Mu Li, Xiangyang Xue
2022 arXiv   pre-print
Extensive experiments demonstrate its validity as a new benchmark for evaluating different object detection methods, and its effectiveness as a pre-training dataset.  ...  Our goal is to simply leverage the training data from existing datasets (LVIS, OpenImages and Object365) with carefully designed principles, and curate a larger dataset for improved detector pre-training  ...  However, MSeg is designed for semantic segmentation and it only contains 200k images over 194 semantics classes.  ... 
arXiv:2203.13249v1 fatcat:hhggcgjfljcmhpzgxjqi4xxu3m

Table of Contents

2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
2866 Xiang Li (HKUST), Tianhan Wei (HKUST), Yau Pun Chen (HKUST), Yu-Wing Tai (Tencent), and Chi-Keung Tang (HKUST)MSeg: A Composite Dataset for Multi-Domain Semantic Segmentation 2876 John Lambert (Intel  ...  Xu (Computational Biology Department, Carnegie Mellon University, USA) FDA: Fourier Domain Adaptation for Semantic Segmentation 4084 Yanchao Yang (UCLA Vision Lab) and Stefano Soatto (UCLA Vision Lab)  ... 
doi:10.1109/cvpr42600.2020.00004 fatcat:c7els2kee5cq7lh6cemeqhdcoa

Investigation of Image Segmentation, Machine Learning and Knowledge-based Expert System Methods in Remote Sensing [article]

Angelos K. Tzotsos, National Technological University Of Athens, National Technological University Of Athens, Δημήτριος Αργιαλάς
The second contribution of this research involved the design and development of a region-based multi-scale segmentation algorithm with the integration of complex texture features.  ...  The implemented algorithm is called MSEG and can be described as a region merging procedure.  ...  Acknowledgements I would like to thank my advisor, Professor Demetre Argialas for his guidance through the years and for his insightful comments and review of this manuscript.  ... 
doi:10.26240/heal.ntua.1644 fatcat:xkmajws4ezflppi53imcstpobm

DEVIANT: Depth EquiVarIAnt NeTwork for Monocular 3D Object Detection [article]

Abhinav Kumar, Garrick Brazil, Enrique Corona, Armin Parchami, Xiaoming Liu
2022 arXiv   pre-print
Even then, all monocular 3D detectors use vanilla blocks to obtain the 3D coordinates, a task for which the vanilla blocks are not designed for.  ...  As a result, DEVIANT is equivariant to the depth translations in the projective manifold whereas vanilla networks are not.  ...  Xuepeng Shi and Li Wang shared details of their cross-dataset evaluation [67] and Waymo experiments [80] respectively. Shengjie Zhu helped us with the monocular depth (BTS) [41] experiments.  ... 
arXiv:2207.10758v1 fatcat:xeu425kukbbs5jdoj4bhgktchm

Enhancing Photorealism Enhancement

Stephan R Richter, Hassan Abu Al Haija, Vladlen Koltun
To address this we propose a new strategy for sampling image patches during training.  ...  We analyze scene layout distributions in commonly used datasets and find that they differ in important ways.  ...  We employ MSeg [74] for semantic segmentation and VGG-16 [95] for perceptual feature extraction.  ... 
doi:10.1109/tpami.2022.3166687 pmid:35412970 fatcat:clwdszmvmbfbpkem3mu6uxedgy

Διερεύνηση Οντολογιών, Μηχανικής Μάθησης, Μορφομετρίας, και Αντικειμενοστρεφούς Ανάλυσης Εικόνας για την Αναγνώριση Γεωμορφών, Κτιρίων, και Μεταβολών Κτιρίων [article]

Αργυρός Αργυρίδης, National Technological University Of Athens, National Technological University Of Athens
., an engineering services company which provided the dataset and reference data to conduct this research.  ...  In Bannour and Hudelot (2014) a multi-stage reasoning approach was developed, to perform semantic annotation/tagging on a target dataset containing ground scenes.  ...  A web server integrating segmentation algorithms such as MSEG (Tzotsos and Argialas 2006) and SPOR for the classification process is required.  ... 
doi:10.26240/heal.ntua.2838 fatcat:nkalagqkjrfyjebkx6h4pfrgwm

The Schulze Method of Voting [article]

Markus Schulze
2022 arXiv   pre-print
Furthermore, we propose a generalization of the Condorcet criterion to multi-winner elections. This paper contains a large number of examples to illustrate the proposed methods.  ...  We propose a new single-winner election method ("Schulze method") and prove that it satisfies many academic criteria (e.g. monotonicity, reversal symmetry, resolvability, independence of clones, Condorcet  ...  Woodall, and Thomas Zavist for fruitful discussions.  ... 
arXiv:1804.02973v11 fatcat:ychf6pfp5nevxovjmqmrng2rmu

A Science Gateway for Molecular Simulations

Sandra Gesing, Peter Kacsuk, Miklos Kozlovsky, Georg Birkenheuer, Dirk Blunk, Sebastian Breuers, Andre Brinkmann, Gregor Fels, Richard Grunzke, Sonja Herres-Pawlis, Jens Krüger, Lars Packschies (+6 others)
We present VisIVO, a powerful environment for exploring highly complex multi-dimensional astrophysical datasets on the grid.  ...  The multi-layer architecture of InSilicoLab supports this domain-specific interaction by introducing a 'scientific' layer, responsible for communication with the users, while a separate layer is reserved  ... 
doi:10.18154/rwth-2015-04977 fatcat:4xsrx6padrdpfflja23ztvudnm