We propose a multi-input multi-output fully convolutional neural network model for MRI synthesis. The model is robust to missing data, as it benefits from, but does not require, additional input modalities. The model is trained end-to-end, and learns to embed all input modalities into a shared modality-invariant latent space. These latent representations are then combined into a single fused representation, which is transformed into the target output modality with a learnt decoder. We avoid the need for curriculum learning by exploiting the fact that the various input modalities are highly correlated. We also show that by incorporating information from segmentation masks the model can both decrease its error and generate data with synthetic lesions. We evaluate our model on the ISLES and BRATS data sets and demonstrate statistically significant improvements over state-of-the-art methods for single input tasks. This improvement increases further when multiple input modalities are used, demonstrating the benefits of learning a common latent space, again resulting in a statistically significant improvement over the current best method. Finally, we demonstrate our approach on non skull-stripped brain images, producing a statistically significant improvement over the previous best method. Code is made publicly available at https://github.com/agis85/multimodal_brain_synthesis.

Graphical AbstractFinely-grained annotated datasets for image-based plant phenotyping Massimo Minervini, Andreas Fischbach, Hanno Scharr, Sotirios A. TsaftarisIn this paper we present a collection of benchmark datasets for the development and evaluation of computer vision and machine learning algorithms in the context of plant phenotyping. We provide annotated imaging data and suggest suitable evaluation criteria for plant/leaf segmentation, detection, tracking as well as classification and regression problems. The Figure symbolically depicts the data available together with ground truth segmentations and further annotations and metadata. Research Highlights• First comprehensive annotated datasets for computer vision tasks in plant phenotyping.• Publicly available data and evaluation criteria for eight challenging tasks.• Tasks include fine-grained categorization of age, developmental stage, and cultivars.• Example test cases and results on plant and leaf-wise segmentation and leaf counting. ABSTRACTImage-based approaches to plant phenotyping are gaining momentum providing fertile ground for several interesting vision tasks where fine-grained categorization is necessary, such as leaf segmentation among a variety of cultivars, and cultivar (or mutant) identification. However, benchmark data focusing on typical imaging situations and vision tasks are still lacking, making it difficult to compare existing methodologies. This paper describes a collection of benchmark datasets of raw and annotated top-view color images of rosette plants. We briefly describe plant material, imaging setup and procedures for different experiments: one with various cultivars of Arabidopsis and one with tobacco undergoing different treatments. We proceed to define a set of computer vision and classification tasks and provide accompanying datasets and annotations based on our raw data. We describe the annotation process performed by experts and discuss appropriate evaluation criteria. We also offer exemplary use cases and results on some tasks obtained with parts of these data. We hope with the release of this rigorous dataset collection to invigorate the development of algorithms in the context of plant phenotyping but also provide new interesting datasets for the general computer vision community to experiment on. Data are publicly available at http://www.plant-phenotyping.org/datasets.

Image-based plant phenotyping is a growing application domain of computer vision in agriculture. A key task is the segmentation of all individual leaves in images. Here we focus on the most common rosette model plants Arabidopsis and young tobacco. Although leaves do share appearance and shape characteristics, the presence of occlusions and variability in leaf shape and pose, as well as imaging conditions, render this problem challenging. The aim of this paper is to compare several leaf segmentation solutions on a unique and first of its kind dataset containing images from typical phenotyping experiments. In particular, we report and discuss methods and findings of a collection of submissions for the first Leaf Segmentation Challenge (LSC) of the Computer Vision Problems in Plant Phenotyping (CVPPP) workshop in 2014. Four methods are presented: three segment leaves via processing the distance transform in an unsupervised fashion, and the other via optimal template selection and Chamfer matching. Overall, we find that although separating plant from background can be achieved with satisfactory accuracy (>90% Dice score), individual leaf segmentation and counting remain challenging when leaves overlap. Besides, accuracy is lower for younger leaves. We find also that variability in datasets does affect outcomes. Our findings motivate further investigations and development of specialized algorithms for this particular application, and that challenges of this form are ideally suited for advancing the state of the art. Data are publicly available (http://www.plantphenotyping.org/CVPPP2014-dataset) to support future challenges beyond segmentation within this application domain.