ACDC: The Adverse Conditions Dataset with Correspondences for Semantic Driving Scene Understanding

By Christos Sakaridis, Dengxin Dai, Luc Van Gool. IEEE International Conference on Computer Vision, 2021

701 citations recorded by the source corpus.

Level 5 autonomy for self-driving cars requires a robust visual perception system that can parse input images under any visual condition. However, existing semantic segmentation datasets are either dominated by images captured under normal conditions or are small in scale. To address this, we introduce ACDC, the Adverse Conditions Dataset with Correspondences for training and testing semantic segmentation methods on adverse visual conditions. ACDC consists of a large set of 4006 images which are equally distributed between four common adverse conditions: fog, nighttime, rain, and snow. Each adverse-condition image comes with a high-quality fine pixel-level semantic annotation, a corresponding image of the same scene taken under normal conditions, and a binary mask that distinguishes between intra-image regions of clear and uncertain semantic content. Thus, ACDC supports both standard semantic segmentation and the newly introduced uncertainty-aware semantic segmentation. A detailed empirical study demonstrates the challenges that the adverse domains of ACDC pose to state-of-the-art supervised and unsupervised approaches and indicates the value of our dataset in steering future progress in the field. Our dataset and benchmark are publicly available.

View Paper: https://www.semanticscholar.org/paper/787b9f2a81412a05cb26269a1e2740c746192f04

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Jiaming Liu
2. Senqiao Yang
3. Peidong Jia
4. Ming Lu
5. Yandong Guo
6. Shanghang Zhang
7. Wei Xue
8. Olga Fink

Citations in the corpus

• Continual Test-Time Domain Adaptation (2022, background): “The target domain contains images from various scenarios from the Adverse Conditions Dataset (ACDC) [50].”
• HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (2022, unknown): “For certain tasks, such as semantic segmentation, providing the annotations is particularly labor-intensive as pixel-wise labels of the entire image are necessary, which can take more than one hour per image [11,51].”
• Toward Fast, Flexible, and Robust Low-Light Image Enhancement (2022, unknown): “As for nighttime semantic segmentation, we utilized 400 images in ACDC [19] for training and the remaining 106 images as the evaluated dataset.”
• ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (2023, methodology): “For segmentation CTTA [59, 61], we evaluate our method on Cityscapes-to-ACDC, where the Cityscapes dataset [12] serves as the source domain, and the ACDC dataset [51] represents the target domains.”