Attribute Error while training in Jeston Orin

Hi Team,
I am using Jeston AGX Orin 64GB developer kit.
image

I have build the cmake and make file. I am facing issue for farthest sample point the error is as mentioned below.
File “train.py”, line 576, in
train()
File “train.py”, line 449, in train
pred, end_points = model.get_model(
File “/home/Documents/test/model.py”, line 36, in get_model
l1_xyz, l1_points, l1_indices = model_sa_module(
File “/home/Documents/test/util/model_util.py”, line 96, in model_sa_module
new_xyz, new_points, idx, grouped_xyz = sample_and_group(npoint, radius, nsample, xyz, points, use_xyz)
File “/home/Documents/test/util/model_util.py”, line 32, in sample_and_group
xyz, farthest_point_sample(npoint, xyz)
File “/home/Documents/test/tf_ops/tf_sampling.py”, line 71, in farthest_point_sample
return sampling_module.farthest_point_sample(inp, npoint)
AttributeError: module ‘5fa89fc3154996733eabb433e18fa62f’ has no attribute ‘farthest_point_sample’

Could you please let me know how to resolve it.

Hi,

There are some dependencies between the GPU driver and the CUDA library.
TensorFlow v1.15.5+nv22.5 is not compatible with the JetPack 5.1.2 environment.

Please install the below package instead:
https://developer.download.nvidia.com/compute/redist/jp/v512/tensorflow/

Thanks.

1 Like

Hi,
I have two jetpacks available in Jeston Orin Jetpack 5.0, and 5.1.2.
For Jeston 5.0, I have installed TensorFlow v1.15.5+nv22.5.
image

Could it be because of these two versions of jetpack.

Hi,

TensorFlow v1.15.5+nv22.5 is not compatible with JetPack 5.1.2.
Please find the above link to install the v2.12.0+nv23.06 and try it again.

Thanks.

1 Like