postdoctoral research assocoate University of Arizona Lawrenceville, Georgia
Body of Abstract: Understanding three-dimensional (3D) root traits is essential to improve the resource uptake of the plant and increase the sequestration of atmospheric carbon into the soil. We developed an image-based 3D root phenotyping platform, Digital Imaging of Root Traits 3D (DIRT/3D) (Liu, Barrow et al. 2021), which can build 3D root models and measure 18 architecture traits from field-grown maize root crowns excavated with the Shovelomics technique. However, the large amount of images needed (~ 3,600) to build a 3D model requires a long scanning time.
Is it possible to use fewer images to build a 3D model by upgrading the 3D reconstruction pipelines? We compared five open-source 3D reconstruction pipelines and their performance and concluded that COLMAP is a solution to significantly reduce the number of images needed. We then compared the correlation of root trait measurements generated by the DIRT/3D using COLMAP-based 3D reconstruction with our previous DIRT/3D pipeline that uses a VisualSFM-based 3D reconstruction. We computed the trait measurements on the previously published dataset of 12 genotypes with 5~10 replicates per genotype(Liu, Paul Bonelli et al. 2022).
The comparison results showed that, 1) the average number of images needed to build a denser 3D model can be reduced from ~3,600 to ~600; 2) the denser 3D models improved the correlation of 3D root trait measurement; 3) reduced number of images helps to reduce data collection time, and computation time if using GPU acceleration. Overall, the updated DIRT/3D 2.0 pipeline enables quicker image collection without compromising the quality of 3D root-trait measurements compared to previous results.
Reference
Liu, S., et al. (2021). "DIRT/3D: 3D root phenotyping for field-grown maize (Zea mays)." Plant Physiology187(2): 739-757.
Liu, S. B., et al. (2022). Comparison of open-source image-based reconstruction pipelines for 3D root phenotyping of field-grown maize. 2022 NAPPN Conference Proceedings.