search

TRI-ML / vidar

Other
573 stars 67 forks source link

configs/papers/packnet/inference_packnet.yaml for kitti_raw #16

Open bayesyf opened 2 years ago

bayesyf commented 2 years ago

|****| | DEPTH | abs_rel | sqr_rel | rmse | rmse_log | silog | a1 | a2 | a3 | |****| | * KITTI_raw-eigen_test_files-velodyne-cam0 | |****| | DEPTH|(0)_0 | 0.926 | 14.268 | 18.905 | 2.857 | 61.165 | 0.000 | 0.000 | 0.000 | | DEPTH|GT(0)_0 | 0.490 | 5.353 | 12.158 | 0.616 | 61.164 | 0.286 | 0.533 | 0.734 | | DEPTH|PP(0)_0 | 0.928 | 14.300 | 18.920 | 2.871 | 60.396 | 0.000 | 0.000 | 0.000 | | DEPTH|PP_GT(0)_0 | 0.474 | 5.093 | 12.128 | 0.610 | 60.396 | 0.290 | 0.539 | 0.742 | |****| | * KITTI_raw-eigen_test_files-groundtruth-cam0 | |****| | DEPTH|(0)_0 | 0.927 | 13.978 | 18.515 | 2.843 | 58.465 | 0.000 | 0.000 | 0.000 | | DEPTH|GT(0)_0 | 0.443 | 4.918 | 11.932 | 0.593 | 58.465 | 0.310 | 0.580 | 0.772 | | DEPTH|PP(0)_0 | 0.929 | 14.011 | 18.530 | 2.857 | 57.563 | 0.000 | 0.000 | 0.000 | | DEPTH|PP_GT(0)_0 | 0.427 | 4.658 | 11.892 | 0.586 | 57.563 | 0.316 | 0.589 | 0.782 |

bayesyf commented 2 years ago

the results are not good, and the used checkpoint has been downloaded from https://tri-ml-public.s3.amazonaws.com/github/packnet-sfm/models/PackNet01_MR_selfsup_K.ckpt , then rename it , am i wrong?

VitorGuizilini-TRI commented 2 years ago

Seems like you haven't loaded your model properly, can you share your config file?

bayesyf commented 2 years ago

Seems like you haven't loaded your model properly, can you share your config file?

I used the default config file, and not revised wrapper: recipe: wrapper|default arch: model: file: depth/SelfSupervisedModel checkpoint: data/vidar/models/papers/final/PackNet_MR_selfsup_KITTI.ckpt networks: depth: recipe: networks/packnet|default depth_range: [0.1,100.0] pose: recipe: networks/conv_pose_net|default evaluation: depth: recipe: evaluation/depth|kitti_resize datasets: validation: recipe: datasets/kitti|validation_mr save: recipe: save|depth_splitname