class_weights

[lifangyan16]

class_weights = [1938651, 1242339, 608870, 1699694, 2794560, 195000, 115990, 549838, 531470, 292971, 196633, 59032, 209046, 39321] in tools.py .
I want to know how these numbers get. Because if I want to use other datasets, I need to know how they get and change them for my dataset. Thank you.

[ooooohhhhhxxx]

I think the numbers in class_weights represent the total amount of points of each class(from the RandLA-Net author replied on RandLA-Net repositories), but what I'm wondering is there only 8 classes in semantic3D (9 classes if count in unlabeled points) and 13 classes in S3DIS, so I don't know why there is 14 weights.
I also want to use my own data to train model, have you done it?

[ooooohhhhhxxx]

class_weights = [1938651, 1242339, 608870, 1699694, 2794560, 195000, 115990, 549838, 531470, 292971, 196633, 59032, 209046, 39321] in tools.py .
I want to know how these numbers get. Because if I want to use other datasets, I need to know how they get and change them for my dataset. Thank you.

Hi, I have changed weight to my dataset(like I said in previous comment, my dataset has 19 classes and I changed weights to the amount of each class), I successfully run the train.py with argument dataset_sampling as naive but when I use active_learning sampling method it occurred an error (cuDNN error：CUDNN_STATUS_EXECUTION_FAILED) and this error not occured when I use semantic3d as dataset, I am still try to find out why.

[A-Chas]

class_weights = [1938651, 1242339, 608870, 1699694, 2794560, 195000, 115990, 549838, 531470, 292971, 196633, 59032, 209046, 39321] in tools.py . I want to know how these numbers get. Because if I want to use other datasets, I need to know how they get and change them for my dataset. Thank you.

Hello, I also have this problem now. My own dataset category is 4, class_ The program can run when the weight is 4, class_ An error will be reported if the weight is 5. However, there will always be a category whose Accuracy and Iou are 0. Please tell me what code needs to be changed if you want to run your own dataset. If we can, we can add a friend to communicate: [email protected]
=== EPOCH 1/6 ===
Training loss: 1.5586808 Validation loss: 2.9165625
Accuracy | 0 | 1 | 2 | 3 | OA
Training: | 0.139 | 0.040 | 0.760 | 0.156 | 0.463
Validation: | 0.000 | 0.000 | 0.991 | 0.009 | 0.576
IoU | 0 | 1 | 2 | 3 | mIoU
Training: | 0.058 | 0.009 | 0.409 | 0.048 | 0.138
Validation: | 0.000 | 0.000 | 0.575 | 0.004 | 0.330
Time elapsed: 8 min 51 s
=== EPOCH 2/6 ===
Training loss: 1.0715053 Validation loss: 1.0353577
Accuracy | 0 | 1 | 2 | 3 | OA
Training: | 0.100 | 0.012 | 0.933 | 0.048 | 0.561
Validation: | 0.850 | 0.000 | 0.990 | 0.000 | 0.696
IoU | 0 | 1 | 2 | 3 | mIoU
Training: | 0.053 | 0.003 | 0.548 | 0.015 | 0.175
Validation: | 0.672 | 0.000 | 0.654 | 0.000 | 0.622
Time elapsed: 8 min 44 s
=== EPOCH 3/6 ===
Training loss: 1.2013940 Validation loss: 2.4099397
Accuracy | 0 | 1 | 2 | 3 | OA
Training: | 0.314 | 0.002 | 0.909 | 0.005 | 0.568
Validation: | 0.000 | 0.000 | 0.983 | 0.000 | 0.564
IoU | 0 | 1 | 2 | 3 | mIoU
Training: | 0.110 | 0.000 | 0.560 | 0.001 | 0.192
Validation: | 0.000 | 0.000 | 0.564 | 0.000 | 0.360

[A-Chas]

I think the numbers in class_weights represent the total amount of points of each class(from the RandLA-Net author replied on RandLA-Net repositories), but what I'm wondering is there only 8 classes in semantic3D (9 classes if count in unlabeled points) and 13 classes in S3DIS, so I don't know why there is 14 weights. I also want to use my own data to train model, have you done it?

Hello, I also have this problem now. My own dataset category is 4, class_ The program can run when the weight is 4, class_ An error will be reported if the weight is 5. However, there will always be a category whose Accuracy and Iou are 0. Please tell me what code needs to be changed if you want to run your own dataset. If we can, we can add a friend to communicate: [email protected]
=== EPOCH 1/6 ===
Training loss: 1.5586808 Validation loss: 2.9165625
Accuracy | 0 | 1 | 2 | 3 | OA
Training: | 0.139 | 0.040 | 0.760 | 0.156 | 0.463
Validation: | 0.000 | 0.000 | 0.991 | 0.009 | 0.576
IoU | 0 | 1 | 2 | 3 | mIoU
Training: | 0.058 | 0.009 | 0.409 | 0.048 | 0.138
Validation: | 0.000 | 0.000 | 0.575 | 0.004 | 0.330
Time elapsed: 8 min 51 s
=== EPOCH 2/6 ===
Training loss: 1.0715053 Validation loss: 1.0353577
Accuracy | 0 | 1 | 2 | 3 | OA
Training: | 0.100 | 0.012 | 0.933 | 0.048 | 0.561
Validation: | 0.850 | 0.000 | 0.990 | 0.000 | 0.696
IoU | 0 | 1 | 2 | 3 | mIoU
Training: | 0.053 | 0.003 | 0.548 | 0.015 | 0.175
Validation: | 0.672 | 0.000 | 0.654 | 0.000 | 0.622
Time elapsed: 8 min 44 s
=== EPOCH 3/6 ===
Training loss: 1.2013940 Validation loss: 2.4099397
Accuracy | 0 | 1 | 2 | 3 | OA
Training: | 0.314 | 0.002 | 0.909 | 0.005 | 0.568
Validation: | 0.000 | 0.000 | 0.983 | 0.000 | 0.564
IoU | 0 | 1 | 2 | 3 | mIoU
Training: | 0.110 | 0.000 | 0.560 | 0.001 | 0.192
Validation: | 0.000 | 0.000 | 0.564 | 0.000 | 0.360

[lifangyan16]

I think the numbers in class_weights represent the total amount of points of each class(from the RandLA-Net author replied on RandLA-Net repositories), but what I'm wondering is there only 8 classes in semantic3D (9 classes if count in unlabeled points) and 13 classes in S3DIS, so I don't know why there is 14 weights. I also want to use my own data to train model, have you done it?

Hello, I also have this problem now. My own dataset category is 4, class_ The program can run when the weight is 4, class_ An error will be reported if the weight is 5. However, there will always be a category whose Accuracy and Iou are 0. Please tell me what code needs to be changed if you want to run your own dataset. If we can, we can add a friend to communicate: [email protected] === EPOCH 1/6 === Training loss: 1.5586808 Validation loss: 2.9165625 Accuracy | 0 | 1 | 2 | 3 | OA Training: | 0.139 | 0.040 | 0.760 | 0.156 | 0.463 Validation: | 0.000 | 0.000 | 0.991 | 0.009 | 0.576 IoU | 0 | 1 | 2 | 3 | mIoU Training: | 0.058 | 0.009 | 0.409 | 0.048 | 0.138 Validation: | 0.000 | 0.000 | 0.575 | 0.004 | 0.330 Time elapsed: 8 min 51 s === EPOCH 2/6 === Training loss: 1.0715053 Validation loss: 1.0353577 Accuracy | 0 | 1 | 2 | 3 | OA Training: | 0.100 | 0.012 | 0.933 | 0.048 | 0.561 Validation: | 0.850 | 0.000 | 0.990 | 0.000 | 0.696 IoU | 0 | 1 | 2 | 3 | mIoU Training: | 0.053 | 0.003 | 0.548 | 0.015 | 0.175 Validation: | 0.672 | 0.000 | 0.654 | 0.000 | 0.622 Time elapsed: 8 min 44 s === EPOCH 3/6 === Training loss: 1.2013940 Validation loss: 2.4099397 Accuracy | 0 | 1 | 2 | 3 | OA Training: | 0.314 | 0.002 | 0.909 | 0.005 | 0.568 Validation: | 0.000 | 0.000 | 0.983 | 0.000 | 0.564 IoU | 0 | 1 | 2 | 3 | mIoU Training: | 0.110 | 0.000 | 0.560 | 0.001 | 0.192 Validation: | 0.000 | 0.000 | 0.564 | 0.000 | 0.360

I suggest you install a Pycharm. These kinds of issues are usually because of dimension of array.