# input
1. dentin, enamel, overlap csv files
2. mapping_rgb.npy
3. thickness.npy
4. front.ply
5. (normal.ply, back.ply)

# resin
1. train on lab rather than rgb
2. no constraint -> scaling of original tooth -> template

# overlapping
1. parallel all points
2. add 縮放: dx -> weighted sum of thickness of NNs

# 誤差可能
1. 照片沒有校正好
2. 原本牙齒的反射/折射沒有考慮進去
3. 單點模擬不足夠（需要考慮鄰近點）
4. 公式和model不足以fit data