Recent advances in semantic segmentation using deep learning methods have achieved promising results on several benchmark datasets. However, the primary challenge involved in such segmentation approaches is the availability of applicable training data. Since only experts are equipped to effectively annotate (or label) any available data for training semantic segmentation networks, the effort and cost involved can be considerable, especially for larger datasets. In this paper, we aim to address this problem by proposing a Thrifty Annotation Generation approach that records high performance on segmentation networks with minimal expert effort and cost (intervention). We present a deep active learning framework that combines the use of marker-controlled watershed (MC-WS) algorithm to generate pseudo labels for segmentation networks (U-Net) and active learning to significantly minimize effort and cost by selecting only the most impactful training data for labeling. We built the initial U-Net model by generating pseudo labels for the training data using MC-WS. We then make use of the uncertainty information (entropy) of each image provided by the U-Net to determine the most uncertain or effective images for expert labeling. We evaluated the TAG approach using the 2012 ISBI Challenge dataset for 2D segmentation and a novel Biofilm dataset. Our approach achieved promising segmentation accuracy (IoU) and classification accuracy with minimal expert intervention. The results of our experiments also indicate that the TAG approach can be generalized to achieve high-performance segmentation results on any dataset using minimal expert effort and cost.