search

dylran / crowddiff

MIT License
104 stars 12 forks source link

Nan value #26

Open BurstLink666 opened 4 months ago

BurstLink666 commented 4 months ago

I got a training result as below:

targets: tensor([[ 80.], [ 71.], [ 65.], [149.]], device='cuda:0'), pred: tensor([[33.4143], [ nan], [32.8215], [32.5122]], device='cuda:0', grad_fn=)

I followed your introduction to do all the preprocess and trained the model with JHU dataset. Is it a random error?

KkkQqq11 commented 3 months ago

I also get this problem， can anyone tell me how to handle this nan question