37,312 Hits in 6.7 sec

Multi-Loss Weighting with Coefficient of Variations [article]

Rick Groenendijk, Sezer Karaoglu, Theo Gevers, Thomas Mensink
2020 arXiv   pre-print
In contrast to many loss weighting methods in literature, we focus on single-task multi-loss problems, such as monocular depth estimation and semantic segmentation, and show that multi-task approaches  ...  The validity of the approach is shown empirically for depth estimation and semantic segmentation on multiple datasets.  ...  The authors of [14, 15] formulate a joint likelihood estimate to determine task weights based on homoscedastic aleatoric uncertainty.  ... 
arXiv:2009.01717v2 fatcat:syli2ro4jjfr5iwmb7tmpdiwp4

SUB-Depth: Self-distillation and Uncertainty Boosting Self-supervised Monocular Depth Estimation [article]

Hang Zhou, Sarah Taylor, David Greenwood
2021 arXiv   pre-print
Additionally, SUB-Depth enables models to estimate uncertainty on depth output.  ...  To take advantage of this multi-task setting, we propose homoscedastic uncertainty formulations for each task to penalize areas likely to be affected by teacher network noise, or violate SDE assumptions  ...  result is a multi-task learning system, which trains Θ depth for an image reconstruction task and a selfdistillation task using the sum of task-dependent uncertainty weighted losses.  ... 
arXiv:2111.09692v2 fatcat:c2w77uk3qbhjfoqddvwgno5yr4

Multitask Learning with Single Gradient Step Update for Task Balancing [article]

Sungjae Lee, Youngdoo Son
2020 arXiv   pre-print
However, learning multiple tasks simultaneously can be more difficult than learning a single task because it can cause imbalance among tasks.  ...  To address the imbalance problem, we propose an algorithm to balance between tasks at the gradient level by applying gradient-based meta-learning to multitask learning.  ...  The architectures of ordinary multitask learning, Uncertainty, GradNorm, Split-only, and the proposed method are based on SegNet [15] .  ... 
arXiv:2005.09910v2 fatcat:qnteww5mzjftbflk7ia6kdkrpa

MultiNet++: Multi-Stream Feature Aggregation and Geometric Loss Strategy for Multi-Task Learning [article]

Sumanth Chennupati, Ganesh Sistu, Senthil Yogamani, Samir A Rawashdeh
2019 arXiv   pre-print
Current work on multi-task learning networks focus on processing a single input image and there is no known implementation of multi-task learning handling a sequence of images.  ...  Multi-task learning is commonly used in autonomous driving for solving various visual perception tasks. It offers significant benefits in terms of both performance and computational complexity.  ...  Aditya Viswanathan and Dr. Thibault Julliand for helpful discussions.  ... 
arXiv:1904.08492v2 fatcat:ob6bo36oardubot35ejs7inabi

CodeMapping: Real-Time Dense Mapping for Sparse SLAM using Compact Scene Representations [article]

Hidenobu Matsuki, Raluca Scona, Jan Czarnowski, Andrew J. Davison
2021 arXiv   pre-print
We build on CodeSLAM and use a variational autoencoder (VAE) which is conditioned on intensity, sparse depth and reprojection error images from sparse SLAM to predict an uncertainty-aware dense depth map  ...  State-of-the-art sparse visual SLAM systems provide accurate and reliable estimates of the camera trajectory and locations of landmarks.  ...  the learning task and can lead to more accurate results.  ... 
arXiv:2107.08994v1 fatcat:suf6x3hqnjgfdbj7awyj6nwr2y

Semi-Supervised Learning with Mutual Distillation for Monocular Depth Estimation [article]

Jongbeom Baek, Gyeongnyeon Kim, Seungryong Kim
2022 arXiv   pre-print
We propose a semi-supervised learning framework for monocular depth estimation.  ...  Compared to existing semi-supervised learning methods, which inherit limitations of both sparse supervised and unsupervised loss functions, we achieve the complementary advantages of both loss functions  ...  This research was supported by the MSIT, Korea (IITP-2022-2020-0-01819, ICT Creative Consilience program), and National Research Foundation of Korea (NRF-2021R1C1C1006897).  ... 
arXiv:2203.09737v1 fatcat:yiobudoacbdtheq6mymnmnkfky

Mono-SF: Multi-View Geometry Meets Single-View Depth for Monocular Scene Flow Estimation of Dynamic Traffic Scenes [article]

Fabian Brickwedde, Steffen Abraham, Rudolf Mester
2019 arXiv   pre-print
Mono-SF jointly estimates the 3D structure and motion of the scene by combining multi-view geometry and single-view depth information.  ...  Mono-SF considers that the scene flow should be consistent in terms of warping the reference image in the consecutive image based on the principles of multi-view geometry.  ...  The methods are divided into four groups: First, multi-task CNNs; second, combining optical flow and single-view depth estimation as individual tasks; third, multi-body or non-rigid SfM-based approaches  ... 
arXiv:1908.06316v1 fatcat:7rn4uahxtvfbbbf7e3nx5fizwi

Multi-Task Learning as Multi-Objective Optimization [article]

Ozan Sener, Vladlen Koltun
2019 arXiv   pre-print
We apply our method to a variety of multi-task deep learning problems including digit classification, scene understanding (joint semantic segmentation, instance segmentation, and depth estimation), and  ...  Our method produces higher-performing models than recent multi-task learning formulations or per-task training.  ...  We directly estimate disparity instead of depth and later convert it to depth using the provided camera intrinsics.  ... 
arXiv:1810.04650v2 fatcat:lc5w3fu5d5epppclysgh5j3qhm

Multi-Task Learning for Scalable and Dense Multi-Layer Bayesian Map Inference [article]

Lu Gan, Youngji Kim, Jessy W. Grizzle, Jeffrey M. Walls, Ayoung Kim, Ryan M. Eustice, Maani Ghaffari
2021 arXiv   pre-print
It removes the need for a robot to access and process information from many separate maps when performing a complex task and benefits from the correlation between map layers, advancing the way robots interact  ...  This paper presents a novel and flexible multi-task multi-layer Bayesian mapping framework with readily extendable attribute layers.  ...  A review of the related work on multi-layer robotic mapping, self-supervised learning for traversability estimation and deep multi-task learning is given in Section II.  ... 
arXiv:2106.14986v1 fatcat:46cq2tsyvfgy7drac2aqi2awwq

BinsFormer: Revisiting Adaptive Bins for Monocular Depth Estimation [article]

Zhenyu Li, Xuyang Wang, Xianming Liu, Junjun Jiang
2022 arXiv   pre-print
It mainly focuses on two crucial components in the specific task: 1) proper generation of adaptive bins and 2) sufficient interaction between probability distribution and bins predictions.  ...  We further integrate a multi-scale decoder structure to achieve a comprehensive understanding of spatial geometry information and estimate depth maps in a coarse-to-fine manner.  ...  To solve the issue, some methods [3, 21] reformulate depth estimation as a per-pixel classification-regression task (Fig. 1c ), learning probabilistic representations on each pixel and predicting the  ... 
arXiv:2204.00987v1 fatcat:vkrchntfo5aa7e3frnmcu4s6gy

Bayesian Deep Neural Networks for Supervised Learning of Single-View Depth [article]

Javier Rodríguez-Puigvert, Rubén Martínez-Cantín, Javier Civera
2021 arXiv   pre-print
Finally, we explore the use of depth uncertainty for pseudo-RGBD ICP and demonstrate its potential to estimate accurate two-view relative motion with the real scale.  ...  In this paper, we evaluate scalable approaches to uncertainty quantification in single-view supervised depth learning, specifically MC dropout and deep ensembles.  ...  Uncertainty is often present in the formulation of model-based estimators (e.g., [1] ), but much less in the training of deep learning models.  ... 
arXiv:2104.14202v3 fatcat:5nnsa5jbn5c37po722kau3y4wm

UM-Adapt: Unsupervised Multi-Task Adaptation Using Adversarial Cross-Task Distillation [article]

Jogendra Nath Kundu, Nishank Lakkakula, R. Venkatesh Babu
2019 arXiv   pre-print
Moreover, the resulting semi-supervised framework outperforms the current fully-supervised multi-task learning state-of-the-art on both NYUD and Cityscapes dataset.  ...  UM-Adapt yields state-of-the-art transfer learning results on ImageNet classification and comparable performance on PASCAL VOC 2007 detection task, even with a smaller backbone-net.  ...  This work was supported by a CSIR Fellowship (Jogendra) and a grant from RBCCPS, IISc. We also thank Google India for the travel grant.  ... 
arXiv:1908.03884v3 fatcat:l7ejobww6jh3xjg5ezkcgt4e6q

Conf-Net: Toward High-Confidence Dense 3D Point-Cloud with Error-Map Prediction [article]

Hamid Hekmatian, Jingfu Jin, Samir Al-Stouhi
2019 arXiv   pre-print
We add an "Error Prediction" unit to our network and present a novel and simple end-to-end method that learns to predict an error-map of depth regression task.  ...  upon the state-of-the-art for monocular depth estimation.  ...  We directly learn both dense depth and its corresponding error-map in a multi-tasking manner. 2D dense depth along with the predicted error-map are used to generate a high-confidence 3D dense point-cloud  ... 
arXiv:1907.10148v3 fatcat:ejdpwrurzjcv7gkyntnvbbcmgu

3D Object Detection from Images for Autonomous Driving: A Survey [article]

Xinzhu Ma, Wanli Ouyang, Andrea Simonelli, Elisa Ricci
2022 arXiv   pre-print
Benefiting from the rapid development of deep learning technologies, image-based 3D detection has achieved remarkable progress.  ...  3D object detection from images, one of the fundamental and challenging problems in autonomous driving, has received increasing attention from both industry and academia in recent years.  ...  Multi-Task Prediction 3D detection as multi-task learning. 3D detection can be seen as a multi-task learning problem because it needs to output the class label, location, dimension, and orientation together  ... 
arXiv:2202.02980v2 fatcat:2hela3uz2bgmtgyzpw5sqslqla

Robust Learning Through Cross-Task Consistency [article]

Amir Zamir, Alexander Sax, Teresa Yeo, Oğuzhan Kar, Nikhil Cheerla, Rohan Suri, Zhangjie Cao, Jitendra Malik, Leonidas Guibas
2020 arXiv   pre-print
The proposed formulation is based on inference-path invariance over a graph of arbitrary tasks.  ...  Visual perception entails solving a wide set of tasks, e.g., object detection, depth estimation, etc.  ...  , we reduce the problem to a 'separable' one, devise a tractable training schedule, and use a 'perceptual loss' based formulation.  ... 
arXiv:2006.04096v1 fatcat:uuk7yaowtfgljk6bcoivo5oi6u
« Previous Showing results 1 — 15 out of 37,312 results