Search before asking

• I have searched the Ultralytics YOLO issues and discussions and found no similar questions.

Question

Hello,
I detect points on the face in Cephalometric Images.
I used Yolov8x-pose and Yolo11x-pose models for this.
Training was done over 1000 images.
Train= 800
Valid= 100
Test= 100

I created the labels myself.
I made the labels according to the following order.
Example Label 1
0 0.683254 0.603351 0.492632 0.599330 0.811962 0.592268 2 0.819617 0.577320 2 0.499522 0.564433 2 0.843541 0.727320 2 ...
and
kpt_shape: [44, 3]

Example Label 2
0 0.683254 0.603351 0.492632 0.599330 0.811962 0.592268 0.819617 0.577320 0.499522 0.564433 0.843541 0.727320...
kpt_shape: [44, 3] or kpt_shape: [44, 2]

Here, when I make Labels like “Example Label 1”
Yolov8 model found 43 points out of 44 points as I wanted.
The yolo11 model found 43 out of 44 points as I wanted.
In both tutorials I find a different point too outlier!

Here when I make the labels like “Example Label 2”
On Yolov8 he found almost none of them in the right places.
Yolo11 training was not done.

Here I would like to ask the following questions;
1-) Why didn't I succeed in all points in “Sample Tag 1”?
2-) In “Sample Tag 2” I deleted the “2” in the txt files, what is the importance of this? Is this wrong?
3-) What is the importance of “2” or “3” in the kpt_shape parameter?

Additional

No response