πŸ“• Discussion on Baseline | Global Wheat Challenge 2021

:rocket: Check out the Baseline here .

If you would like to discuss any queries or issues with the baseline, do post them here! You can also share any new interesting ideas you might be trying out!

I noticed in the Rules that β€œThe participant agrees to exclusively use the Global Wheat Head Dataset for the purpose of participating in this competition.”. What does β€œexclusively use” mean in this context? Since it is declared in the next sentence that one is allowed to use Global Wheat Head Dataset outside of this challenge, I assume, it means no external data use in the challenge, correct? Then I noticed use of pretrained model looking through baseline. If everything up to this point is correct, you should define datasets that pretraining is allowed on.

Thanks for noticing this point in the rules.

We will clarify them but:

  • Use of pretrained models on ImageNet / MS COCO are authorized (for instance all models pretrained on timm etc…).
  • Training model on private data / manually labeled data is NOT authorized.

I hope this clarify the rules !

Best,

Etienne

1 Like

Hello,

Is it fine to use yolov5 ( not MIT license) ?

Cheers,

1 Like

YES :slight_smile: ! Every open-source model are authorized even yolo v5 (GPL licence ) !

1 Like

Where can I find the labels csv file?

The labels csv files is contained with the train dataset !

In train.zip, there is train.csv . I think it is the label file.

How can I save the downloaded dataset in google drive?
I am running the baseline model in colab and I do not want to download the dataset again when I want to run the model again after runtime disconnection. I just want to mount drive and load the dataset.

Hi @poojamalagund,

Yes, you can always mount your Google Drive to Colab, here is the example on IO thing from Colab:

https://colab.research.google.com/notebooks/io.ipynb

Edit: Forgot to add link.

1 Like

While running the baseline model in Colab, is there any way that I can utilize the dataset for training without actually downloading it?
Can I load the dataet directly from a link in colab?

No, you need to download, even if you use them on the fly, it will actually download.
So, I don’t think there will be any savings by not downloading? :sweat_smile:

You can use your Google Drive to skip downloading again and again at max.
Let me know in case I understood the question wrongly. :slight_smile:

1 Like

Yes, Thank you :grinning: