A substantial number of prevalent traffic datasets capture a bias towards having more clear and standard driving scenes. Although some recent datasets

have been collected to tackle the issue of the long tail end of the traffic data distribution, still these datasets are not comprehensive enough to cover the various sub-domains of adverse illumination and weather conditions since it

is a resource exhaustive process. Data augmentation is often used as a strategy to improve the diversity of training data for machine learning systems.

While standard augmentation techniques (such as translation and flipping)

help neural networks to generalize over spatial transformations, more nuanced techniques would be required to capture semantically different variations in data. We propose a new data augmentation method that relies on

the use of attribute-conditioned generative models to modify the semantic properties of existing training data. We show that such data augmentation improves the generalization capability of deep networks by analyzing their performance on datasets of traffic objects that are captured (i) at different times of the day and (ii) across different weather conditions.