search

XuyangBai / D3Feat

[TensorFlow] Official implementation of CVPR'20 oral paper - D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features https://arxiv.org/abs/2003.03164
MIT License
259 stars 38 forks source link

IndexError:run demo with 'kitti' model #10

Closed tracykim closed 4 years ago

tracykim commented 4 years ago

HI, i want to run demo with 'kitti' model, so i change the path to 'results_kitti/Log_11011605/'. Howerver, an error occurred on this line IndexError: tuple index out of range I would appreciate it if you could help me.

XuyangBai commented 4 years ago

Hi @tracykim The key 'trans' is only for KITTI dataset, and for the demo there is no trans for the input list. So you may manually set config.dataset = '' after loading the config to avoid this error.

BTW, If you are testing the KITTI model on point cloud of outdoor scenes, then it should work well. But if you use are using the two demo .ply in demo_data selected from 3DMatch, you should take care of the voxel size as well as the receptive field because they have different scales with KITTI dataset. Basically you need to

  1. set the config.first_subsampling_dl to a proper value (e.g. 0.03 for point cloud from 3DMatch. The origin value for KITTI is 0.30 which is too big for point cloud from 3DMatch)
  2. change the scale of the kernel points as I mentioned here.

Hope that helps.

tracykim commented 4 years ago

Hi, thanks for your advice.I will test on my data of outdoor scenes.

tracykim commented 4 years ago

Hi @XuyangBai The performance of D3Feat on my data is well, thanks for your work and help. best.

XuyangBai commented 4 years ago

Hi @tracykim Glad to here that!