<p>Fruit detection, shape measurement, and counting are integral to yield estimation. Detection and classification of individual partial and full leaves from the bunch of leaves are fundamental parameters for various measurements and research in agriculture, forestry, and environmental science. In this paper, we propose deep learning models to detect and classify the shape (partial, full) of individual leaves from bunch leaf images and to count individual fruits from bunch fruit images as they are used in various measurements in agriculture and forestry. Previous studies have yet to experiment simultaneously on highly dense leaf and fruit datasets with different lighting settings for smart agricultural applications. We deploy two light-weighted object detection CNN models (YOlOv8, EfficientDet-D0) and a popular object detection CNN model (Faster R-CNN) for leaf detection and yield estimation. Moreover, no real-time suitable dataset containing highly dense leaves or fruits is available. Therefore, we developed two real-time datasets: one consisting of bunches leaves of nine species (612 images) and the other of papaya fruits (1840 images), collected from Bangladeshi agricultural fields. The datasets contain exceedingly overlapped objects with diverse backgrounds (ordinary and complex) and lighting conditions (sunny, cloudy, and gloomy). From the experimental results, we see that YOLOv8 outperforms other models, achieving 93.88%, 91.59%, and 96.78% precision, recall, and mAP, respectively on the leaf dataset for detecting and classifying individual partial and full leaves from the bunch of leaves, and 99.19%, 98.92%, and 99.43% precision, recall and mAP, respectively on the papaya dataset for detecting individual papaya fruit from the bunch.</p>