Assertion `p_in_data[idx] >= 0 && p_in_data[idx] < depth` failed.

[pobi123]

bug描述 Describe the Bug

I got an error while using paddleseg for semantic segmentation training. Error: C:\home\workspace\Paddle\paddle\phi\kernels\gpu\one_hot_kernel.cu:37 Assertion p_in_data[idx] >= 0 && p_in_data[idx] < depth failed. Illegal index value, Input(input) value should be greater than or equal to 0, and less than depth [20], but received [20].

My dataset has 7910 images, using grayscale annotations, with a total of 20 labels (including background classes).

Here is my environment configuration: cuda 11.6 cudnn 8.4.0 python 3.8.0 paddlapddle 2.5.2 paddleseg 2.9

Can someone help me?

其他补充信息 Additional Supplementary Information

No response

[BiynXu]

Can you please provide reproduction scripts?

[pobi123]

Can you please provide reproduction scripts?

thank you. I've solved the problem. It is caused by the change of pixel value of some pixel points caused by interpolation algorithm during image geometry transformation.