Filters








32,483 Hits in 6.8 sec

Modeling Camera Effects to Improve Visual Learning from Synthetic Data [article]

Alexandra Carlson, Katherine A. Skinner, Ram Vasudevan, Matthew Johnson-Roberson
2018 arXiv   pre-print
Recent work has focused on generating synthetic imagery to increase the size and variability of training data for learning visual tasks in urban scenes.  ...  Sensor effects can degrade real images, limiting generalizability of network performance on visual tasks trained on synthetic data and tested in real environments.  ...  Effects to Improve Deep Vision for Real and Synthetic Data.  ... 
arXiv:1803.07721v6 fatcat:nuzuml5ucnfs3lvm3b3nuyigka

Modeling Camera Effects to Improve Visual Learning from Synthetic Data [chapter]

Alexandra Carlson, Katherine A. Skinner, Ram Vasudevan, Matthew Johnson-Roberson
2019 Lecture Notes in Computer Science  
Recent work has focused on generating synthetic imagery to increase the size and variability of training data for learning visual tasks in urban scenes.  ...  Sensor effects can degrade real images, limiting generalizability of network performance on visual tasks trained on synthetic data and tested in real environments.  ...  However, there still is an absence in the literature examining how to improve failure modes due to sensor effects for learned visual tasks in the wild.  ... 
doi:10.1007/978-3-030-11009-3_31 fatcat:esjapivlefarfkni65xkzsoa4e

Synthesize and Segment: Towards Improved Catheter Segmentation via Adversarial Augmentation

Ihsan Ullah, Philip Chikontwe, Hongsoo Choi, Chang-Hwan Yoon, Sang Hyun Park
2021 Applied Sciences  
We show that an adversarially learned image-to-image translation network can synthesize catheters in X-ray fluoroscopy enabling data augmentation in order to alleviate a low data regime.  ...  In the case of deep learning based methods, the demand for large amounts of labeled data further impedes successful application.  ...  We evaluate four segmentation models to prove the effectiveness of the synthetically generated data, and found that all the models improved performance over models trained without synthetic data.  ... 
doi:10.3390/app11041638 fatcat:e7wewdu2yvekzmvg6n4sfm6bca

Instance-Guided Context Rendering for Cross-Domain Person Re-Identification

Yanbei Chen, Xiatian Zhu, Shaogang Gong
2019 2019 IEEE/CVF International Conference on Computer Vision (ICCV)  
Extensive experiments on Market-1501, DukeMTMC-reID and CUHK03 benchmarks show that the re-id performance can be significantly improved when using our synthetic data in cross-domain re-id model learning  ...  Unlike previous image synthesis methods that transform the source person images into limited fixed target styles, our approach produces more visually plausible, and diverse synthetic training data.  ...  data for re-id model learning.  ... 
doi:10.1109/iccv.2019.00032 dblp:conf/iccv/ChenZG19 fatcat:2ehie77suzbnjpa7hhgzwwwcce

Image-to-Image Translation of Synthetic Samples for Rare Classes [article]

Edoardo Lanzini, Sara Beery
2021 arXiv   pre-print
We explore the use of image-to-image translation methods to close the domain gap between synthetic and real imagery for animal species classification in data collected from camera traps: motion-activated  ...  Learning from few examples is a known challenge for deep learning based classification algorithms, and is the focus of the field of low-shot learning.  ...  Acknowledgements We would like to thank the USGS and NPS for providing data and Microsoft AI for Earth for providing compute resources.  ... 
arXiv:2106.12212v1 fatcat:logankwygvcurkfh6eeybyatoq

How to Train a CAT: Learning Canonical Appearance Transformations for Direct Visual Localization Under Illumination Change

Lee Clement, Jonathan Kelly
2018 IEEE Robotics and Automation Letters  
We further provide a preliminary investigation of transfer learning from synthetic to real environments in a localization context.  ...  We validate our method in multiple environments and illumination conditions using high-fidelity synthetic RGB-D datasets, and integrate the trained models into a direct visual localization pipeline, yielding  ...  However, there has been little work on using machine learning techniques to generate such models from data.  ... 
doi:10.1109/lra.2018.2799741 dblp:journals/ral/ClementK18 fatcat:oiikdehmazcc7gpch6gxyqfb3i

Estimation of Absolute Scale in Monocular SLAM Using Synthetic Data [article]

Danila Rukhovich, Daniel Mouritzen, Ralf Kaestner, Martin Rufli, Alexander Velizhev
2019 arXiv   pre-print
In addition, we exploit a possibility to train the neural network only with synthetic data derived from a computer graphics simulator.  ...  These estimates would improve the overall performance of classical (not deep) SLAM systems and allow metric feature locations to be recovered from a single monocular camera.  ...  To answer this question, one could apply domain adaptation methods, which help improve the visual realism of synthetic data and train the scale estimator with more realistic synthetic data.  ... 
arXiv:1909.00713v1 fatcat:evsntppcybbqjpku2y72l4u5wi

Domain Adaptation through Synthesis for Unsupervised Person Re-identification [article]

Slawomir Bak, Peter Carr, Jean-Francois Lalonde
2018 arXiv   pre-print
As a result, a trained model requires fine-tuning to become effective under an unseen illumination condition.  ...  To achieve better accuracy in unseen illumination conditions we propose a novel domain adaptation technique that takes advantage of our synthetic data and performs fine-tuning in a completely unsupervised  ...  In this case, adding synthetic data improved performance from 7% to 15%. Our domain adaptation technique boosts the performance to 43.0% rank1-accuracy.  ... 
arXiv:1804.10094v1 fatcat:lenvoegct5bqfa554drwv7strq

Domain Adaptation Through Synthesis for Unsupervised Person Re-identification [chapter]

Sławomir Bąk, Peter Carr, Jean-François Lalonde
2018 Lecture Notes in Computer Science  
As a result, a trained model requires fine-tuning to become effective under an unseen illumination condition.  ...  To achieve better accuracy in unseen illumination conditions we propose a novel domain adaptation technique that takes advantage of our synthetic data and performs fine-tuning in a completely unsupervised  ...  In this case, adding synthetic data improved performance from 7% to 15%. Our domain adaptation technique boosts the performance to 43.0% rank1-accuracy.  ... 
doi:10.1007/978-3-030-01261-8_12 fatcat:fdpyp2dh7vf2jfrzzwduifbeee

Visual Compiler: Synthesizing a Scene-Specific Pedestrian Detector and Pose Estimator [article]

Namhoon Lee, Xinshuo Weng, Vishnu Naresh Boddeti, Yu Zhang, Fares Beainy, Kris Kitani, Takeo Kanade
2016 arXiv   pre-print
Given a single image and auxiliary scene information in the form of camera parameters and geometric layout of the scene, the Visual Compiler first infers geometrically and photometrically accurate images  ...  We demonstrate that when real human annotated data is scarce or non-existent, our data generation strategy can provide an excellent solution for bootstrapping human detection and pose estimation.  ...  Learning the Network from Synthetic Data Using the scene-specific data generated above, the visual compiler now generates a visual program, in the form of deep neural network, trained to operate according  ... 
arXiv:1612.05234v1 fatcat:iwg7mrpt7jcjpe6wh5lsxtav2i

Learning Camera-Aware Noise Models [article]

Ke-Chi Chang, Ren Wang, Hung-Jin Lin, Yu-Lun Liu, Chia-Ping Chen, Yu-Lin Chang, Hwann-Tzong Chen
2020 arXiv   pre-print
To tackle this issue, we propose a data-driven approach, where a generative noise model is learned from real-world noise.  ...  The proposed noise model is camera-aware, that is, different noise characteristics of different camera sensors can be learned simultaneously, and a single learned noise model can generate different noise  ...  The triplet loss is essential to learn effective camera-specific latent vectors, and the KL divergence can be significantly reduced from 0.01412 to 0.00159.  ... 
arXiv:2008.09370v1 fatcat:qslln3yofbhyfp5jk3mrxb2doi

Sensor Transfer: Learning Optimal Sensor Effect Image Augmentation for Sim-to-Real Domain Adaptation [article]

Alexandra Carlson, Katherine A. Skinner, Ram Vasudevan, Matthew Johnson-Roberson
2019 arXiv   pre-print
The camera or sensor used to capture a dataset introduces artifacts into the image data that are unique to the sensor model, suggesting that sensor effects may also contribute to domain shift.  ...  Our proposed augmentation pipeline transfers specific effects of the sensor model -- chromatic aberration, blur, exposure, noise, and color temperature -- from a real dataset to a synthetic dataset.  ...  METHODS The objective of the sensor transfer network is to learn the the optimal set of augmentations that transfer sensor effects from a real dataset to a synthetic dataset.  ... 
arXiv:1809.06256v2 fatcat:kblrvcs5cfhh5ovs2ohszodo2u

RSA: Randomized Simulation as Augmentation for Robust Human Action Recognition [article]

Yi Zhang, Xinyue Wei, Weichao Qiu, Zihao Xiao, Gregory D. Hager and Alan Yuille
2019 arXiv   pre-print
In this paper, we propose the Randomized Simulation as Augmentation (RSA) framework which augments real-world training data with synthetic data to improve the robustness of action recognition networks.  ...  An alternative is to make use of simulation data, where all of these factors can be artificially controlled.  ...  We visualize the feature distributions of data from both domains learned by different models in 2D space using t-SNE [20] .  ... 
arXiv:1912.01180v1 fatcat:4g5bxwil4vcpjhxn3sc4sanwcy

Drone Model Identification by Convolutional Neural Network from Video Stream

Mariusz Wisniewski, Zeeshan A. Rana, Ivan Petrunin
2021 2021 IEEE/AIAA 40th Digital Avionics Systems Conference (DASC)  
This synthetic dataset is used to train a convolutional neural network to identify the drone model: DJI Phantom, DJI Mavic, or DJI Inspire.  ...  To achieve this, we show a method of generating synthetic drone images. To create a diverse dataset, the simulation parameters (such as drone textures, lighting, and orientation) are randomized.  ...  For the purposes of testing our model, we are only interested in daytime videos from the visual camera.  ... 
doi:10.1109/dasc52595.2021.9594392 fatcat:4tii4kmlyjah7pyru3eqwklxxa

Large Scale Vehicle Re-Identification by Knowledge Transfer from Simulated Data and Temporal Attention

Viktor Eckstein, Arne Schumann, Andreas Specker
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)  
However, with a growing number of available datasets and well crafted deep learning models, much progress has been made.  ...  Our models, code, and simulated data is available at https://github.com/corner100/ 2020-aicitychallenge-IOSB-VeRi.  ...  We thus aim to use the synthetic data to train a view regression model.  ... 
doi:10.1109/cvprw50498.2020.00316 dblp:conf/cvpr/EcksteinSS20 fatcat:itjldeltezhbxmkdh5ixmbgkna
« Previous Showing results 1 — 15 out of 32,483 results