Not able to import pytorch_lightning in colab

[benihime91]

Not able to import pytorch_lightning in colab

[benihime91]

Hmmm.... that's weird ... @alankritmishra Did you restart your runtime after installing pytorch-lightning ?

[alankritmishra]

Yes a couple of times, still didn't work. I am doing this experiment with my custom dataset. In the requirements you mentioned the version should not be more than 1.0.0. So, i didn't upgrade.

[benihime91]

You could probably try looking here for some solutions https://stackoverflow.com/questions/66538407/unable-to-import-pytorch-lightning-on-google-colab. It's seems its due to a version mismatch between torchtext and pytorch-lightning

[benihime91]

Colab has probably upgraded the default torch and torchvision version.

torch>=1.6.0 torchvision>=0.7.0

You could try reinstalling the versions mentioned above. Or another way would be to uninstall torchtext.

[alankritmishra]

Okay let me try

[benihime91]

@alankritmishra I have updated the demo notebook. Should work on colab now. Do check it out.

[alankritmishra]

Thank you @benihime91 pytorch_lightning imported successfully after uninstalling torchtext . However, it shows there on module called 'model'

when I am trying from model import RetinaNetModel

[alankritmishra]

Thank you @benihime91 pytorch_lightning imported successfully after uninstalling torchtext . However, it shows there on module called 'model'

when I am trying from model import RetinaNetModel

Should we have to track it as another issue?

[benihime91]

Did you try the new demo.ipynb notebook @alankritmishra ?

[alankritmishra]

Yes, and it worked, but I already have csv files for annonation and was trying your original documentation in github.

Here is my Colab notebook for your reference. Thanks again for such quick response.

[benihime91]

Hmmm.... You need to do

import sys
sys.path.append("/content/pytorch_retinanet/")

After cloning the repo and before importing modules from the RetinaNet repo itself.

Also you probably do not need to do

pip install -r /content/pytorch_retinanet/requirements.txt

in colab because these lines are already doing that

!pip install -Uqq pytorch-lightning==1.0.0 omegaconf
!pip install -Uqq git+https://github.com/albumentations-team/albumentations
!pip uninstall torchtext
!git clone https://github.com/benihime91/pytorch_retinanet.git

[alankritmishra]

I came through that, Now I just have to mention the right path of my data in hparams.yaml.

It shows the below error:
FileNotFoundError: [Errno 2] No such file or directory: '/train/_annotations.csv'

Although, I already provide the root path in the root_dir placeholder

[alankritmishra]

Is there a way for the model to retrieve the data and annotation directly from my drive, or do I have to upload it on the runtime environment?

[benihime91]

You need provide the absolute path to the annotations. And root_dir is needed only for COCO datasets

[alankritmishra]

Hi @benihime91 the original issue is resolved so closing this tracker. Can I dm you for this to work?

[alankritmishra]

Hi @benihime91 the original issue is resolved so closing this tracker. Can I dm you for this to work?

I have emailed you.