gtsfm icon indicating copy to clipboard operation
gtsfm copied to clipboard
verified publisher icon borglab

Use MST initialization with huber loss

Open ayushbaid opened this issue 1 year ago • 6 comments

ayushbaid avatar May 18 '24 04:05 ayushbaid

This seems to do better

ba ra

ayushbaid avatar May 18 '24 07:05 ayushbaid

@ayushbaid the diff is against "MST w/o Huber", not "no MST" right? Are we still seeing regressions when compared to "no MST"?

akshay-krishnan avatar May 18 '24 15:05 akshay-krishnan

@ayushbaid the diff is against "MST w/o Huber", not "no MST" right? Are we still seeing regressions when compared to "no MST"?

The metrics are compared against master though

ayushbaid avatar May 18 '24 16:05 ayushbaid

@ayushbaid the diff is against "MST w/o Huber", not "no MST" right? Are we still seeing regressions when compared to "no MST"?

The metrics are compared against master though

Yes, I meant the diff in metrics.

akshay-krishnan avatar May 20 '24 00:05 akshay-krishnan

Also, are you sure inlier weighting was turned on it master when you compared? I vaguely remember @akshay-krishnan accidently turning it off in master.

travisdriver avatar Jun 14 '24 22:06 travisdriver

Why did you turn off inlier weighting?

It was just for experimentation and comparing different sets of values.

Inlier weighting is on right now on master.

ayushbaid avatar Jun 15 '24 23:06 ayushbaid