About meta_info

[CharlesHuan]

Dear authors,

thanks for your great work!

I was trying to make my own custom object to track and was following your tutorial.，I have some questions :
my custom meta_info.txt like this:

It's based on

the visualizations I got weren't good enough,like this:

but directly download the reference images processed by you at here (mouse_processed.tar.gz) ,could get a good visualizations ,like this:

so I guess the problem maybe appear in my meta info.txt
I need some help T-T

[omarirfa]

Hi,

You may have already seen this issue but just wanted to confirm if you also looked at this issue regarding metainfo and followed similar steps from there to create the metainfo file? #7

[CharlesHuan]

yes，I look at this issue regarding metainfo and followed similar steps from there to create my metainfo file #7 X+ direction for the object is(2.271052 0.032156 -0.296206)).like this

and z-(visualize the normal and use the normal as the Z direction,The normal is shown in the left-bottom panel. Note that the normal is opposite to the Z+ direction (up), so the Z+ direction for the object is (-0.0992386 -0.573665 -0.813056)),like this

my meta_info like this(It's also shown in the description, but just to make sure I follow #7):

I have the following files, which are generated by above steps.

and I predict poses for the test video by these command:

I don't know exactly where it went wrong.T-T

[omarirfa]

Strange I also get different results but it is not similar at all to the original one for the mouse. Could you post your segmented mouse i.e. your object_point_cloud.ply? I think @liuyuan-pal can best comment on this since I have also encountered this similar issue while using the code. I am guessing its how the segmentation is being done based on this issue #7 last reply but I am unsure.

Images_Inter:

Object_point_cloud.ply:

Original object_point_cloud.ply:

As you can see the original is very accurately segmented such that there is no aruco board I am guessing its because of the segmentation?

[CharlesHuan]

Strange I think my object_point_cloud.ply is accurate，like this：

[omarirfa]

Oh you didnt segment it by yourself? this is the original file i believe. Unsure tbh what the problem is then

[CharlesHuan]

Oh you didnt segment it by yourself? this is the original file i believe. Unsure tbh what the problem is then

This is actually my own segmented image T-T, because the segmentation is pretty accurate, except for Z + and X +, these are visually very similar.

[liuyuan-pal]

Hi, @CharlesHuan it seems that your object_point_cloud.ply has some problem so that the generated reference images are not centered on the object: (since we use the object_point_cloud.ply to specify the object (size and center) you want to track.)

The reference images should look like:

You may open both the pointcloud.ply and object_point_cloud.ply in the same CloudCompare windows to see if they are aligned or not.

[CharlesHuan]

Hi, @CharlesHuan it seems that your object_point_cloud.ply has some problem so that the generated reference images are not centered on the object: (since we use the object_point_cloud.ply to specify the object (size and center) you want to track.) The reference images should look like:

You may open both the pointcloud.ply and object_point_cloud.ply in the same CloudCompare windows to see if they are aligned or not.

Thanks for your reply！Through your prompts，I found problems in generating database：
It should be like this

but dense clouds generated by the GUI I used,maybe cause the database to be misaligned with the generated cloud.
The problem has now been solved,thanks again！