utiasSTARS / learned_scale_recovery

Accompanying code for 'Self-Supervised Scale Recovery for Monocular Visual Odometry'
47 stars 12 forks source link

Training the Plane Segmentation Network before training depth and ego-motion networks? #1

Open sunnyHelen opened 3 years ago

sunnyHelen commented 3 years ago

Hi. Thanks for sharing your great work. Should we train the Plane Segmentation Network before training the depth and ego-motion networks?

brandonwagstaff commented 3 years ago

Hi there,

There is a three-stage process to training. First, an unscaled depth network is required, which can be trained using our training procedure without using the scale recovery losses. Then, you can use this (unscaled) depth network to train the plane segmentation network. Following this, use the plane segmentation network (with the scale recovery losses) to train a new depth and egomotion network that will be metrically scaled.

sunnyHelen commented 3 years ago

Got it. Thanks a lot for your reply. BTW, do you have the test results of pose on Odometry sequences?