The existing datasets lack the diversity required to train the model so that it performs equally well in real fields  under varying environmental conditions. To address this limitation, we propose to collect a small number of in-field data and use the GAN to generate synthetic data for training the deep learning network. To demonstrate the proposed method, a maize dataset 'IIITDMJ_Maize'  was collected using a drone camera under different weather conditions, including both sunny and cloudy days. The recorded video was processed to sample image frames that were later resized to 224 x 224.

The dataset comprises diverse objects detectable by drones during aerial surveys, encapsulating an extensive array of environmental and man-made elements. Encompassing natural entities like trees, water bodies, terrain features, and vegetation, it also incorporates urban objects such as buildings, roads, vehicles, and infrastructure. The dataset delineates distinct categories, encompassing fine-grained details within each classification, catering to the nuances of aerial detection.

Scene text detection and recognition have attracted much attention in recent years because of their potential applications. Detecting and recognizing texts in images may suffer from scene complexity and text variations. Some of these problematic cases are included in popular benchmark datasets, but only to a limited extent. In this work, we investigate the problem of scene text detection and recognition in a domain with extreme challenges.

Volkswagen Group of America Innovation and Engineering Center California (VW IECC) is a research facility in Belmont, California working on the future of the mobility. In the recent years exciting developments have happened for the autonomous vehicles. In general, lack of data is the main problem to tackle to solve the task of autonomous driving. One of the important tasks in this topic is the overtaking and lane changes, especially in the highway scenarios.