ABSTRACT
This chapter introduces an approach for the estimation of 3D branch (stem) structures of plants from multiview images using a deep learning-aided 3D reconstruction technique. It reviews the prior literature for 3D modeling of plant shoots and image-to-image (I2I) translation, the key component of our 3D reconstruction technique. I2I translation aims at transferring contextual or physical variations between the source and target images. Some traditional I2I translation techniques have been referred to as texture synthesis or texture transfer when the approach focused on the textures. More recent I2I translation methods benefit from deep learning, such as convolutional neural networks with encoder-decoder architectures or generative adversarial networks. The probabilistic 3D branch structure can be converted to an explicit representation of 3D branch paths that can be used for structural analysis applications (e.g., counting branches and measuring their lengths) for plant phenotyping.
