I am unable to understand the significance of iters1 and iters2 in droid_frontend.py even after going through the paper and code. I understand that the update operation does. However I am not clear why two such update sequences are needed.
Specifically, what do the two update operations do in the following code in droid_frontend.py (lines 50 to 66):
` for itr in range(self.iters1):
self.graph.**update**(None, None, use_inactive=True)
# set initial pose for next frame
poses = SE3(self.video.poses)
d = self.video.distance([self.t1-3], [self.t1-2], beta=self.beta, bidirectional=True)
if d.item() < self.keyframe_thresh:
self.graph.rm_keyframe(self.t1 - 2)
with self.video.get_lock():
self.video.counter.value -= 1
self.t1 -= 1
else:
for itr in range(self.iters2):
self.graph.**update**(None, None, use_inactive=True)
# set pose for next itration
self.video.poses[self.t1] = self.video.poses[self.t1-1]
self.video.disps[self.t1] = self.video.disps[self.t1-1].mean()`
Also what does it mean by set pose for next iteration?
I am unable to understand the significance of iters1 and iters2 in droid_frontend.py even after going through the paper and code. I understand that the update operation does. However I am not clear why two such update sequences are needed. Specifically, what do the two update operations do in the following code in droid_frontend.py (lines 50 to 66):
Also what does it mean by set pose for next iteration?