Icevision: AttributeError: module 'mantisshrimp.tfms' has no attribute 'A'

[tugot17]

Code:

from mantisshrimp.all import *

presize = 512
size = 384

train_tfms = tfms.A.Adapter([*tfms.A.aug_tfms(size=size, presize=presize), tfms.A.Normalize()])
valid_tfms = tfms.A.Adapter([*tfms.A.resize_and_pad(size=size), tfms.A.Normalize()])

Throws: AttributeError: module 'mantisshrimp.tfms' has no attribute 'A'.

It is based on the example from docs, is it changed somehow now?

[lgvaz]

That error means you don't have albumentations installed.

For fixing this immediate error you should do pip install albumentations, but consider installing mantis with [all] so problems like this don't happen =)

[tugot17]

I've done as u recommended. It has fixed tfms error however two new problems occured

1) During first run in jupyter after executting from mantisshrimp.all import * I get ModuleNotFoundError: No module named 'mantisshrimp.models.rcnn.fastai' error. Surprisingly when I execute the cell once again the problem disappears (the cell runs normally)

2) There is also a problem with show_record method. It seems as though it ins't included in mantisshrimp.all since when I try to use it I get NameError: name 'show_record' is not defined, any ideas what might be causing that?

[lgvaz]

I think I might have broken something in the latest commit... I'll check it out and let you know :smile:

[lgvaz]

This should be fixed with #321.

It was a very stupid mistake I did in a commit from yesterday, I also updated the tests to pick errors like these from the future, hopefully, you didn't lose much time :sweat_smile:

For updating, you just need to do:

pip install git+git://github.com/airctic/mantisshrimp.git#egg=mantisshrimp[all] --upgrade

---

On a side note, we're already working on a Pypi release, that will give users more stability compared to working directly from the master branch, keep an eye on our community for updates :smile:

[tugot17]

Yeah it fixed my issue, so I was able to finally train the model, thanks for fast help.
I found the new bug (error when resize_and_pad goes beyond the detection), however I think it needs a new issue.

I close this one.