Embedded Files
Challenge
Challenge
The challenge is to perform semantic segmentation on images from the India Driving Dataset. You will be required to create a model that predicts the correct per pixel label for a sample input image which is imperative for any autonomous driving system for its decision making capabilities. The ground truth labels will be provided in the dataset thus making this problem a supervised learning problem and your system will be evaluated based on how close your model's predictions are to the ground truth for the test set.
The challenge is to perform semantic segmentation on images from the India Driving Dataset. You will be required to create a model that predicts the correct per pixel label for a sample input image which is imperative for any autonomous driving system for its decision making capabilities. The ground truth labels will be provided in the dataset thus making this problem a supervised learning problem and your system will be evaluated based on how close your model's predictions are to the ground truth for the test set.
The training dataset can be downloaded from here. Download the IDD - Segmentation (IDD 20k Part II) (5.5 GB) which contains10,000 images and fine semantic segmentation annotation; released in 2018.
The training dataset can be downloaded from here. Download the IDD - Segmentation (IDD 20k Part II) (5.5 GB) which contains10,000 images and fine semantic segmentation annotation; released in 2018.
What's semantic segmentation ?
What's semantic segmentation ?
Semantic segmentation or image segmentation refers the problem of assigning per pixel label in an image keeping in mind the context and semantic information in the image. Semantic segmentation is widely used in different areas and has now become one of the core driving forces of advances in autonomous driving.
Semantic segmentation or image segmentation refers the problem of assigning per pixel label in an image keeping in mind the context and semantic information in the image. Semantic segmentation is widely used in different areas and has now become one of the core driving forces of advances in autonomous driving.
Semantic segmentation is a per-pixel classification problem. Your task is to assign the correct class label for a particular pixel to ensure that all the pixels belonging to a given object are identified as being belonging to that particular object class. There are different evaluation metrics for comprehending the performance of such a system such as the IoU(intersection over union).
Semantic segmentation is a per-pixel classification problem. Your task is to assign the correct class label for a particular pixel to ensure that all the pixels belonging to a given object are identified as being belonging to that particular object class. There are different evaluation metrics for comprehending the performance of such a system such as the IoU(intersection over union).
To know more you can read this amazing blog to get started.
To know more you can read this amazing blog to get started.
Role of Semantic Segmentation in Autonomous Driving
Role of Semantic Segmentation in Autonomous Driving
Semantic segmentation plays a vital role in autonomous driving system since it plays the responsibility of providing the necessary context for understanding the given scene which is then further used to deduce its actions at a given point of time. The quality of the semantic segmentation drives the level of precision to which the system can take its decision. The labels provided are further used to understand how and which objects are in the environment to drive what interaction needs to take place. Any error in semantic segmentation could cost loss of life thus making it imperative that the quality of such a system be high. In the context of Indian roads the IDD provides a unique challenge. While participating you will appreciate the uniqueness of dataset and the challenges posed by the task.
Semantic segmentation plays a vital role in autonomous driving system since it plays the responsibility of providing the necessary context for understanding the given scene which is then further used to deduce its actions at a given point of time. The quality of the semantic segmentation drives the level of precision to which the system can take its decision. The labels provided are further used to understand how and which objects are in the environment to drive what interaction needs to take place. Any error in semantic segmentation could cost loss of life thus making it imperative that the quality of such a system be high. In the context of Indian roads the IDD provides a unique challenge. While participating you will appreciate the uniqueness of dataset and the challenges posed by the task.
Additional Resources
Additional Resources
Blogs
A 2020 guide to Semantic Segmentation [Link]
How to do Semantic Segmentation using Deep learning [Link]
Semantic Segmentation: Introduction to the Deep Learning Technique Behind Google Pixel’s Camera! [Link]
Image Segmentation in 2020: Architectures, Losses, Datasets, and Frameworks [Link]
Semantic Segmentation on Cityscapes test [Link]
Research Papers
Zhang, Hang, et al. “ResNeSt: Split-Attention Networks.” ArXiv:2004.08955 [Cs], Apr. 2020. arXiv.org, https://arxiv.org/pdf/2004.08955v1.pdf
Cheng, Bowen, et al. “Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation.” ArXiv:1911.10194 [Cs], Mar. 2020. arXiv.org, https://arxiv.org/pdf/1911.10194v3.pdf
Wang, Jingdong, et al. “Deep High-Resolution Representation Learning for Visual Recognition.” ArXiv:1908.07919 [Cs], Mar. 2020. arXiv.org, https://arxiv.org/pdf/1908.07919v2.pdf
Yuan, Yuhui, et al. “Object-Contextual Representations for Semantic Segmentation.” ArXiv:1909.11065 [Cs], July 2020. arXiv.org, https://arxiv.org/pdf/1909.11065v5.pdf
Chen, Liang-Chieh, et al. “Rethinking Atrous Convolution for Semantic Image Segmentation.” ArXiv:1706.05587 [Cs], Dec. 2017. arXiv.org, https://arxiv.org/pdf/1706.05587v3.pdf
Page updated
Report abuse