search

ClementPinard / SfmLearner-Pytorch

Pytorch version of SfmLearner from Tinghui Zhou et al.
MIT License
1.01k stars 224 forks source link

Seq 10 is not in the test_secens.txt #157

Closed AlbertOh90 closed 4 months ago

AlbertOh90 commented 5 months ago

Hello Clement!

Thanks for sharing your work. Just find out that there is a mapping between the KITTI raw data and KITTI odometry data mapping = { "2011_10_03_drive_0027": "00", "2011_10_03_drive_0042": "01", "2011_10_03_drive_0034": "02", "2011_09_26_drive_0067": "03", "2011_09_30_drive_0016": "04", "2011_09_30_drive_0018": "05", "2011_09_30_drive_0020": "06", "2011_09_30_drive_0027": "07", "2011_09_30_drive_0028": "08", "2011_09_30_drive_0033": "09", "2011_09_30_drive_0034": "10" } (https://openaccess.thecvf.com/content_CVPR_2019/supplemental/Chen_Selective_Sensor_Fusion_CVPR_2019_supplemental.pdf) However, the sequence 09 and 10 are not listed in the test_scences.txt. Does it mean that we should not use 09 and 10 for evaluation if the model is trained on the KITTI_raw_data that is prepared by the provided command.

Thanks!

ClementPinard commented 4 months ago

Hello,

yes Indeed. If you want to test odometry, you need to remove sequence 09 and 10 from training.

I did not spend too much time on this matter, and someone else pointed it out in the past : #67

I anyone wants to provide a training recipe especially for odometry and share the weights, they're welcome to do so ! :smile:

AlbertOh90 commented 4 months ago

I did not check the previous questions. Got it and Thank you for your quick reply!