Custom Data Training on Colab

[TalonBvV]

Has anyone successfully done detection training on colab?

I am looking for a version of DocTR that outputs the progress in a more simple way, just basic standard output (print).

The reason for this is due to the fact that I haven't succeeded in training on colab due to the library used to output progress (fastprogress), and if it is possible to get it working it would still be tedious.

[odulcy-mindee]

Hello @TalonBvV 👋,

Can you describe what's the problem you got with fastprogress on Colab please ?

[TalonBvV]

Sorry, there was a typo in that notebook, please see the correct one below:
https://colab.research.google.com/drive/1zZDif8X15L24VUAp199_iDKSyjTqZQtu?usp=sharing

[odulcy-mindee]

@TalonBvV No problem. I've been able to reproduce this issue. I'm able to make the progress bar using tqdm instead of fastprogress. I'll create a new branch / PR, I'll ping you when it's ready

[TalonBvV]

You are a legend!!!

[odulcy-mindee]

@TalonBvV Can you try to use this branch please ? odulcy-mindee:replace_fastprogress_with_tqdm

[TalonBvV]

Works perfectly! thanks a million, I should be able to get my proof of concept done now.

To anyone who may have the same problem, if the branch has not been merged yet you can clone it like below:
git clone https://github.com/odulcy-mindee/doctr.git --branch replace_fastprogress_with_tqdm