Visualization of Part Segmentation predictions

[Ritchizh]

Hi! I'm following your 3D Shape Part Segmentation work. Could you please provide instuctions how to visualize predictions like in the paper:

In main.py test() we get predictions for batches: https://github.com/CVMI-Lab/PAConv/blob/696d36844e6b2f475f732579c24f67ae8059449d/part_seg/main.py#L370-L371 How to combine them into predictions for an individual test file?

[mutianxu]

Hi,

I use MatLab to visualize, you can also choose other tools.

For the progress:

1. Store the input 3d points (size bn3);
2. Store the predicted output (size bn50);
3. "project" the output into 3d points, where each point corresponds to a label. Each label will have different colors.

[Ritchizh]

@mutianxu thank you for the reply!

1. b - is the number of batches in 1 epoch? n - number of points in 1 batch?
2. As I understand, 3 in input are RGB, 50 in output are labels. pred_choice = seg_pred.data.max(1)[1] will produce output of bn1, points coordinates should be the same as in input - what do you mean by projection?

[mutianxu]

@mutianxu thank you for the reply!

1. b - is the number of batches in 1 epoch? n - number of points in 1 batch?
2. As I understand, 3 in input are RGB, 50 in output are labels. pred_choice = seg_pred.data.max(1)[1] will produce output of bn1, points coordinates should be the same as in input - what do you mean by projection?

1. Yes
2. Projection means you use bn3 as the 3d coordinates of points, and use bn1 as the color of each point.