nianticlabs / simplerecon

[ECCV 2022] SimpleRecon: 3D Reconstruction Without 3D Convolutions
Other
1.31k stars 121 forks source link

generate_test_tuples.py with scannet format #38

Open yohanlegars opened 1 year ago

yohanlegars commented 1 year ago

I tested the generate_test_tuples.py for SceneNet dataset (i converted it to the scannet dataset format so that i can use the scannet dataset class). you suggest that frame-**.pose.txt must follow this form: -0.384739 0.271466 -0.882203 4.98152 0.921157 0.0521417 -0.385682 1.46821 -0.0587002 -0.961035 -0.270124 1.51837 a (3,4) matrix. But I am getting the following error:

scannet_dataset.py", line 579, in load_pose cam_T_world = np.linalg.inv(world_T_cam) numpy.linalg.LinAlgError: Last 2 dimensions of the array must be square

which to me indicates that frame-*****.pose.txt must be a square matrix (4,4). Let me know if it makes sense.

Furthermore, do you think testing simplerecon on SceneNet dataset (input size: (320,240)) following Scannet dataset format is a reasonable approach? Best, Yohan

Morpheus3000 commented 1 year ago

The transformation matrix needs to be a 4x4. So you need to add a [0, 0, 0, 1] row to your poses.