Vinicius F. Arruda, Thiago M. Paixão, Rodrigo F. Berriel, Alberto F. De Souza, Claudine Badue, Nicu Sebe and Thiago Oliveira-Santos
Published in IJCNN 2019 Conference: 10.1109/IJCNN.2019.8852008
The preprint version can be accessed here.
Deep learning techniques have enabled the emergence of state-of-the-art models to address object detection tasks. However, these techniques are data-driven, delegating the accuracy to the training dataset which must resemble the images in the target task. The acquisition of a dataset involves annotating images, an arduous and expensive process, generally requiring time and manual effort. Thus, a challenging scenario arises when the target domain of application has no annotated dataset available, making tasks in such situation to lean on a training dataset of a different domain. Sharing this issue, object detection is a vital task for autonomous vehicles where the large amount of driving scenarios yields several domains of application requiring annotated data for the training process. In this work, a method for training a car detection system with annotated data from a source domain (day images) without requiring the image annotations of the target domain (night images) is presented. For that, a model based on Generative Adversarial Networks (GANs) is explored to enable the generation of an artificial dataset with its respective annotations. The artificial dataset (fake dataset) is created translating images from day-time domain to night-time domain. The fake dataset, which comprises annotated images of only the target domain (night images), is then used to train the car detector model. Experimental results showed that the proposed method achieved significant and consistent improvements, including the increasing by more than 10% of the detection performance when compared to the training with only the available annotated data (i.e., day images).
The source code used for the CycleGAN model was made publicly available by Van Huy.
The source code used for the Faster R-CNN model was made publicly available by Xinlei Chen.
For training the Faster R-CNN, a pre-trained resnet-101 model was used to initializate the process an can be downloaded here.
The trained model used in this paper is available here.
The trained models used in this paper are available here.
Download the Berkeley Deep Drive dataset here. It is only necessary to download the Images and Labels files.
After downloading the BDD dataset, the Images and Labels will be placed into the zipped files bdd100k_images.zip and bdd100k_labels.zip respectively. In the same directory, place the provided source code filter_dataset.py from this repository with the folder lists.
On the terminal, run: python filter_dataset.py.
It will take a few minutes, and at the end, the folder images and labels will contain the images and bounding boxes of the images respectively.
Available here.
Videos demonstrating the inference performed by the trained Faster R-CNN model which yielded the best results with our proposed system.
| Testing on Day+Night Dataset | Testing on Night Dataset |
|---|---|
 Inferences performed on day+night dataset |
 Inferences performed on night dataset |
@inproceedings{arruda2019ijcnn,
author={Vinicius F. Arruda, Thiago M. Paixão, Rodrigo F. Berriel, Alberto F. De Souza, Claudine Badue, Nicu Sebe and Thiago Oliveira-Santos},
booktitle={2019 International Joint Conference on Neural Networks (IJCNN)},
title={Cross-Domain Car Detection Using Unsupervised Image-to-Image Translation: From Day to Night},
year={2019},
doi={10.1109/IJCNN.2019.8852008},
ISSN={2161-4407},
month={July}
}