DIGIT error

[andy8902]

When I use this web(jetson-inference/digits-native.md at master · dusty-nv/jetson-inference · GitHub) to start the DIGITS Server, the error is on the below picture.
It appears “ImportError: No module named requests”. How to solve it?

[dusty_nv]

Hi @andy8902, requests is a Python pip package that is from the DIGITS requirements.txt list:

NVIDIA/DIGITS/blob/0100f5604c1938cab1c8577af40219b80a7e5775/requirements.txt#L6

It should have been installed when you did the sudo pip install -r $DIGITS_ROOT/requirements.txt step. Did you do that, or were there errors?

Note that support for DIGITS in the jetson-inference tutorial is deprecated - it is recommended to migrate to the PyTorch tutorial here: GitHub - dusty-nv/jetson-inference: Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson.

[andy8902]

I am very sorry to ask a DIGIT problem. Do the DIGIT support for python2? Can the DIGIT support for python3? I have installed pytorch on Jetson nano. I used this web(jetson-inference/pytorch-collect.md at master · dusty-nv/jetson-inference · GitHub) to train googlenet network.But this method can’t set network parameters and architecture. Except DIGITS software can change the neural network structure and parameters, other things can be changed? Because I am a master student, I will write the paper to analysis network.

Thanks for your help.

[AastaLLL]

Hi,

DIGITs is tested and published by python2.

You can try to set the parameter and architecture with pyTorch frameworks directly.
Some detail you can find in the pyTorch document. For example:
https://pytorch.org/tutorials/beginner/finetuning_torchvision_models_tutorial.html

Thanks.

[andy8902]

When I use DIGIT to classify a single picture, I don’t know how to solve it. This is error’s information:

Tensorflow support disabled.
2020-07-29 21:43:38 [ERROR] IOError:
Traceback (most recent call last):
File “/home/bmw/DIGITS/digits/tools/inference.py”, line 293, in
args[‘resize’]
File “/home/bmw/DIGITS/digits/tools/inference.py”, line 56, in infer
model = Job.load(model_dir)
File “/home/bmw/DIGITS/digits/job.py”, line 37, in load
with open(filename, ‘rb’) as savefile:
IOError: [Errno 2] No such file or directory: ‘/home/bmw/DIGITS/digits/jobs/20200729-213949-3a04/status.pickle’

[AastaLLL]

Hi,

The status.pickle should be generated when creating the database.
Could you check if this folder do present in your environment first?

/home/bmw/DIGITS/digits/jobs/20200729-213949-3a04/status.pickle

Thanks.

[andy8902]

[AastaLLL]

Hi,

The filename is changed from status.pickle into status.pickle.tmp.

Here are two suggestions for you:
1. Rename the status.pickle.tmp back to status.pickle to see if works.
2. Create a new task with the latest model as pre-train, then the status.pickle should be generated again.

Thanks.

[andy8902]

I have a error on Digits run:

Tensorflow support disabled.
2020-08-14 11:05:38 [ERROR] EOFError:
Traceback (most recent call last):
File “/home/bmw/DIGITS/digits/tools/inference.py”, line 293, in
args[‘resize’]
File “/home/bmw/DIGITS/digits/tools/inference.py”, line 56, in infer
model = Job.load(model_dir)
File “/home/bmw/DIGITS/digits/job.py”, line 38, in load
job = pickle.load(savefile)
File “/usr/lib/python2.7/pickle.py”, line 1384, in load
return Unpickler(file).load()
File “/usr/lib/python2.7/pickle.py”, line 864, in load
dispatchkey
File “/usr/lib/python2.7/pickle.py”, line 886, in load_eof
raise EOFError
EOFError

How to solve it?

[andy8902]

When I use Nvidia DIGITs to train googlenet with caffe, how can I do to train high accuracy and low loss’s network which is network?

[AastaLLL]

Hi,

You can try the different learning rate and gradient descent approach to see if helps.

Thanks.

[andy8902]

Hello, I use Nvidia DIGIT web to train googlenet network, how can I solve overfitting’s problem ?

[AastaLLL]

Hi,

A simple way is to add more dataset or stop the training earlier.

Thanks.