zju3dv / manhattan_sdf

Code for "Neural 3D Scene Reconstruction with the Manhattan-world Assumption" CVPR 2022 Oral
https://zju3dv.github.io/manhattan_sdf/
Other
491 stars 35 forks source link

The coordinate of pose.txt? #32

Closed mtk380 closed 1 year ago

mtk380 commented 1 year ago

I have read the data preperation guid. You said you get the pose from clomap, which means x points right, y points down, z points towards scene. If i use the blender data (x points right, y points up, z points towards scene), should i transform the pose to match yours like this? ` # coordinate change

location = np.array([bcam.matrix_world.decompose()[0]]).T

R_bcam2world = np.array(bcam.matrix_world.decompose()[1].to_matrix())

R_col2world = np.matmul(R_col2bcam, R_bcam2world)

T_col2world = np.matmul(R_col2bcam,location)

extrinsic = np.concatenate((R_col2world, T_col2world), axis=1)`
ghy0324 commented 1 year ago

We use the pose provided by the datasets, which use the same definition of coordinate as colmap. And yes, you need to transform If you use blender data.