In this method, we use the visible and the amodal masks from ORCNN to estimate the diameter. From the contour of the amodal masks, an enclosing circle is fitted. Then, a pixel-to-mm conversion factor is calculated from the filtered XYZ-coordinates of the visible mask (using histogram filtering). The real-world diameter is estimated by multiplying the diameter of the enclosing circle (in pixels) with the pixel-to-mm conversion factor.
To use this method, please follow this procedure:
- Annotate the dataset, see ANNOTATE.md
- Train ORCNN, see ORCNN_Train_and_Evaluate_AmodalVisibleMasks.ipynb
- Estimate the diameter from the visible and the amodal masks, see Diameter_estimation_AmodalVisibleMasks.ipynb