airctic / icevision

An Agnostic Computer Vision Framework - Pluggable to any Training Library: Fastai, Pytorch-Lightning with more to come
https://airctic.github.io/icevision/
Apache License 2.0
847 stars 150 forks source link

Importing from the old library #434

Closed muellerzr closed 4 years ago

muellerzr commented 4 years ago

🐛 Bug

Describe the bug Currently imports.py calls import fastai2.vision.all. fastai2 is no longer the home, it's fastai. As a result the API is a bit behind, leading to a few other bugs further down the pipeline (because it's searching fastai2, not fastai)

To Reproduce Steps to reproduce the behavior:

  1. pip install fastai --upgrade
  2. from icevision.all import *

Expected behavior No errors being run

Desktop (please complete the following information):

lgvaz commented 4 years ago

Good catch!

Currently the supported versions for fastai are ">= 0.0.18, <0.0.26", we do need to upgrade that together with fastcore.

I'll leave this issue open until it's solved

lgvaz commented 4 years ago

This is also a good first issue for starting contributing to the library, I'll tag it so anyone interested can pick it up =)

FraPochetti commented 4 years ago

Let me take a stab to this one. I should have some time today.

lgvaz commented 4 years ago

It will be necessary to update both fastai and fastcore at the same time (because they co-depend), the first error you encounter will probably be related to the Tuple to fasttuple renaming.

Anyways, if you have any problems feel free to share it here, glhf!

muellerzr commented 4 years ago

Another good way to see what all the changes were is to take a look at this thread, there's also tips for automating the changes: https://forums.fast.ai/t/major-new-changes-and-features/40742

bguan commented 4 years ago

Hi, playing w IceVision and ran into this issue #434. Anyone working on this? I can take it on if nobody is working on it.

FraPochetti commented 4 years ago

Hi @bguan I was supposed to look into it yesterday but I am having unexpected personal commitments which will keep me busy for the next couple or days. Please feel free to go ahead and grab it. Just let me know if you do so, to avoid stepping onto each other's toes.

bguan commented 4 years ago

@FraPochetti I am happy to take a stab, i think its mostly done, just tracking down 2 test failures from FastAI's learner.

bguan commented 4 years ago

I submitted a PR for this issue, please review at your earliest convenience. https://github.com/airctic/icevision/pull/454

lgvaz commented 4 years ago

closed with #454