anhttran
commented
3 years ago Dear,
The evaluation code is very messy and was never prepared for public release. I uploaded some parts of my code & I hope you find them useful: https://drive.google.com/drive/folders/1KbkaDBkopkPI_eyFJstLaPWhyIQVFwJC
Precisely, we did not use the ground-truth models as is. Instead, we aligned the ground-truth models and cropped them. You can find the ground-truth models we used in AlignedModels.zip**. You can see that each subject has 3 models:
- XX.obj: the aligned model
- XX_n.obj: the aligned model with normals
- XX_crop.ply: the cropped model
Matlab code: matlab_code.zip The main file is compareModelCNN_MICC.m** It calls 3 external programs to do ICP alignment, to compute 3D-RMSE, and to compute depth-based error measures.
Visual C++ code: Match3D.zip**
- ICP alignment: use icp.cpp
- 3D-RMSE computation: use compareModels.cpp
The code for depth-based error measures is a bit complicated since it requires rendering. You can find a code segment in depth_based_pseudo.txt**.
Best regards, Anh
Hi! Thanks for your great work in 3D face reconstruction!
I have a question about the processing of MICC Florence dataset.
How did you process the scan results, especially the crop step? Because I find it difficult to find the nose point accurately T_T...