A Paired Domain Adaptation Dataset for Clear and Snowy Weather: CADC+

CADC+ is the first paired weather domain adaptation dataset for autonomous driving in winter conditions. CADC+ extends the Canadian Adverse Driving Conditions dataset (CADC) using clear weather data that was recorded on the same roads and in the same period as CADC. 

To create CADC+, we pair each CADC sequence with a clear weather sequence that matches the snowy sequence as closely as possible. CADC+ thus minimizes the domain shift resulting from factors unrelated to the presence of snow. This pairing enables, for the first time, domain adaptation experiments and cross-evaluation of methods across both snowy and clear conditions. CADC+ enables the development of more snow-invariant 3D object detection methods and more realistic style transfer methods such as de-snowing or snowfall simulation. We call our clear weather extension CADC-clear.

CADC+ features 74 sequence pairs from snowy and clear weather, with each sequence consisting of at least 100 frames. 3D bounding box annotations are provided for CADC-clear's full validation set and for every 10th frame of its training sequences. While this labelling strategy stemmed from a budget constraint, our prior experience with sparse labelling shows that it is possible to train a model using only sparse labels and perform closely to that of one trained on a fully human-annotated dataset. While CADC-clear's training set is sparsely labelled, we still fully label its validation set to ensure accurate evaluation.