No grab available

[Ruangq]

I have a problem. When I use the model I trained to predict, the success rate of grabbing is very low, about 0.003.
I hope you can give me some help to solve this problem.

[Ruangq]

My environment is tensorflow2.4，cuda11.1, python3.7

[Martin Sundermeyer]

Do you run your training on a Nvidia 30XX GPU? In that case, you might try the environment posted in #19

[Xiaolin Lin]

@Ruangq Hi, you can try my environment file in #19. In my case is that TensorFlow 2.4 and 2.6 didn't work on my laptop. only TensorFlow 2.5 works well with cuda11.2 /11.3 on my laptop. Btw, don't use conda to install TensorFlow.

[Ruangq]

@xlim1996
Hello, During my training, the following situation occurred.
But the training can be carried out normally and the loss is reduced normally.
my GPU is A40.

[Ruangq]

Will this affect the training results and how to solve this problem? @xlim1996
I used the environment file of #19.

[Xiaolin Lin]

@Ruangq
hi,
I didn't retrain the model. So I'm unable to answer your question.
But the environment file works well for using contact graspnet. You may need to ask others who have experience of the retraining model.