| List |
| Topic: | Adaptace obrazových dat z denní do noční domény s kontrolou osvětlení |
|---|---|
| Department: | Katedra kybernetiky |
| Supervisor: | Ing. David Hurych, Ph.D. |
| Announce as: | Diplomová práce, Semestrální projekt |
| Description: | Deep learning techniques have enabled the emergence of several state-of-the-art models to address problems in different domains, such as image classification, regression, object detection and semantic segmentation [6, 7]. However, these techniques are data-driven, which means that the performance achieved in a test dataset strongly depends on the quality of the training dataset. Therefore, the lack of annotated datasets may hinder the training of these models. A challenging scenario arises when a high-performing model in one domain (i.e., target domain) is desired, but the model is trained on a distinct, yet analogous, domain (i.e., source domain). In these situations, the target domain and the source domain are very close in semantics, but are very different in appearance.
Capturing enough variety in the night scenes is very challenging because the amount of pedestrians and other classes instances is significantly reduced. In this work the student shall focus on translating images from day as the source domain into night as the target domain in driving scenarios. There are several methods [1, 2, 3, 4] that already tried to deal with this problem. E.g. [2, 3] use a variant of cycleGAN [5] to generate night-time images from day-time images in order to train a functional night-time lane [3] or car [2] detectors. One of the main problems of methods based on cycleGAN, which do not focus on keeping the scene segmentation intact, is the loss of instances from less frequent object classes. |
| Bibliography: | [1] GAN-Based Day-to-Night Image Style Transfer for Nighttime Vehicle Detection
Che-Tsung Lin, Sheng-Wei Huang, Yen-Yi Wu, Shang-Hong Lai IEEE Transactions on Intelligent Transportation Systems, 2020 [2] Cross-Domain Car Detection Using Unsupervised Image-to-Image Translation: From Day to Night Vinicius F. Arruda, Thiago M. Paixao, Rodrigo Ferreira Berriel, Alberto F. De Souza, Claudine Badue, Nicu Sebe, Thiago Oliveira-Santos International Joint Conference on Neural Networks, 2019 [3] Lane Detection in Low-light Conditions Using an Efficient Data Enhancement: Light Conditions Style Transfer Tong Liu, Zhaowei Chen, Yi Yang, Zehao Wu, Haowei Li Computing Research Repository, 2020 [4] ForkGAN: Seeing into the Rainy Night Ziqiang Zheng, Yang Wu,, Xinran Han and Jianbo Shi IEEE European Conference on Computer Vision, 2020 [5] Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks Jun-Yan Zhu, Taesung Park, Phillip Isola, Alexei A. Efros IEEE International Conference on Computer Vision, 2017 [6] Mask R-CNN Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross B. Girshick IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020 [7] High-Level Semantic Feature Detection: New Perspective for Pedestrian Detection Wei Liu,Shengcai Liao, Weiqiang Ren, Weidong Hu, Yinan Yu IEEE Computer Vision and Pattern Recognition, 2019 |