ICLR Workshop Challenge #1: CGIAR Computer Vision for Crop Disease
Helping East Africa
$5 000 USD
Completed (almost 6 years ago)
Classification
Computer Vision
1050 joined
338 active
Start
Jan 29, 20
Close
Mar 28, 20
Reveal
Mar 29, 20
[CLOSED] pretrained weights are/aren't external data
Help · 18 Feb 2020, 15:17 · edited ~2 hours later · 6
At the first glance, the rules seemed to be clear about the use of external data "You may use only the datasets provided for this competition." Now, I have some doubts I would like to clarify .
Is this rule only regarding the extra images and are the publicly available pretrained weights allowed?
Update: External images are not allowed. Opensoursed pretrained weights are allowed.
In the 'possibility of external data usage' topic they said "You may use transfer learning as long as it is an open-source package and does not require any external data sets."
Good question, I am waiting for the answer too. I also think that pretrained model using transfer learning is based on external data so must not be used here
So if I want to use some external data, as long as I train a model with that data and open source that model, I'm good?
I assume they wanted to say open sourced weights obtained from a predefined open sourced dataset such as imagenet, cocco, openimages, mnist etc
A whitelist of allowed datasets for pretrained models makes sense.
e.g. only imagenet
I want @Zindi to clarify this