Annotated examples for the OrienText300k dataset

[moneyturtle123]

Thank you for open-source your work!
After reading the paper and related code, can you provide a few annotated examples and display demos for the OrienText300k dataset?

[qizekun]

Hello, thank you for your suggestion.
We add some examples of OrienText300K in the readme, using Cap3D's point cloud and visualization.

[moneyturtle123]

While reproducing the code, I wanted to visually inspect the direction vector predicted by the PointSo model on the original point cloud. I used init_position and the predicted direction vector init_orientation to compute:
second_position = init_position + init_orientation * scale，scale=0.1

Then, I connected init_position and second_position to represent the predicted direction vector of the object. As shown in the figure, the red line on the cup's point cloud represents the model-predicted direction vector, while the blue line represents the target direction vector. However, from the figure, it is clear that the result is incorrect. Repeating the process multiple times yields the same issue.

I would like to ask whether I have misunderstood the model-predicted direction vector.

[moneyturtle123]

The Task is "upright the water cup"

[qizekun]

Hello, you can try using base_finetune.pth as the PointSO model. In addition, PointSO does not input complete sentences; it inputs open Vocabulary words, which for this task might be "bottle cap" or "top". And ensure the input points is object point clouds.

[youngfriday]

@moneyturtle123 I have the same problem with you. Have you found any solution for that? Thx!

[Litetras]

when i use 'cut' for a knife, i also found the orientation is incorrect , is it because of the training dataset?