Googlenet on jetson nano

andy8902 · May 12, 2020, 2:40am

How can I decrese overfitting like Weight Decay ,Early Stopping and Dropout?

AastaLLL · May 12, 2020, 6:38am

Hi,

This is a training stage issue.

Usually, we don’t apply training on the Jetson platform due to some hardware design limitation.
It’s recommended to apply training on our desktop GPU instead.

To solve the over-fitting issue, you can try GooglenetV2 which introduces a batch normalization layer.
Thanks.

andy8902 · May 12, 2020, 12:37pm

I use this web(jetson-inference/pytorch-collect.md at master · dusty-nv/jetson-inference · GitHub) to train googlenet. Sorry, I am not clear to say this problem.Can I adjust googlenet’s program function on train ? Or does this web(jetson-inference/pytorch-collect.md at master · dusty-nv/jetson-inference · GitHub) have the googlenet’s program to adjust googlenet’s program function on train ? I say program function like this.

AastaLLL · May 25, 2020, 7:04am

Hi,

Sorry for the late update.

The tutorial is based on pyTorch frameworks so you can set the training parameter as pyTorch manner.

github.com

dusty-nv/pytorch-classification/blob/master/train.py#L35


      
          import torchvision.transforms as transforms
          import torchvision.datasets as datasets
          import torchvision.models as models
          
          
from torch.utils.tensorboard import SummaryWriter
          
          
from reshape import reshape_model
          
          
model_names = sorted(name for name in models.__dict__
              if name.islower() and not name.startswith("__")
              and callable(models.__dict__[name]))
          
          
#
          # parse command-line arguments
          #
          parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
          
          
parser.add_argument('data', metavar='DIR',
                              help='path to dataset')
          parser.add_argument('--model-dir', type=str, default='models', 
          				help='path to desired output directory for saving model '

Ex.

python3 train.py ... --epochs 100 ...

Thanks.