alibaba
About the Camera Embedding
Great Work!
I want to fine-tune this model with my own dataset. Could you please provide some information about the camera embeddings? Thanks for your time again!
Lines 22-23 was aimed to convert cameras from OpenGL to OpenCV coodinates. Blender render objs in OpenGL coordinates. Again the camera matrix here is world2camera transformation. You could use np.linalg.inv(camera_matrix) to convert it to camera2world matrix. A standard process of using your own dataset will be using the rendering script from Objaverse Dataset to render you 3D models and directly load the save json files of it using the read_camera_matrix_single function. To summary, differences in camera lies in CV&&GL coordinates and c2w&&w2c transformation. Try to figure out what you camera matrix is and apply corresponding transformation is enough.
