3. “CNNs have been quickly adopted by the
computer vision community”
“We predict a paradigm shift in image-based
phenotyping thanks to deep learning approaches"
4.
5.
6. Published Oct 2016: “method outperforms recent approaches on multiple person segmentation,
and all state of the art approaches on the Plant Phenotyping dataset for leaf counting.”
7.
8. Xiaodan Liang and (2015). Proposal-free Network for Instance-level Object Segmentation. CoRR, abs/1509.02636.
9. Yi Li and (2016). Fully Convolutional Instance-aware Semantic Segmentation.
Editor's Notes
2500 dataset
accuracy: root tips, 98%, shoots 97% Other “recent state-of-the-art” performed 80-90% accuracy on [2] leafs and [11] phenotypic profiling in time-lapse microscopy
CNN uses all pixels as features rather than manually crafting features, for example areas of high contrast such as edges and corners.
2500 dataset
accuracy: root tips, 98%, shoots 97% Other “recent state-of-the-art” performed 80-90% accuracy on [2] leafs and [11] phenotypic profiling in time-lapse microscopy
CNN uses all pixels as features rather than manually crafting features, for example areas of high contrast such as edges and corners.
2500 dataset
accuracy: root tips, 98%, shoots 97% Other “recent state-of-the-art” performed 80-90% accuracy on [2] leafs and [11] phenotypic profiling in time-lapse microscopy
CNN uses all pixels as features rather than manually crafting features, for example areas of high contrast such as edges and corners.