office-31 demo doesn't work about libmtl HOT 6 CLOSED

The environment we used is as follows:

• Python == 3.8.3
• torch ==1.8.1+cu111
• torchvision == 0.9.1+cu111

and it can work.

The version of pip is not the same version published on github

Yes. The code on github is lastest. We will release the lastest version on PyPi soon.

from libmtl.

Thanks，I have a another question:
The framework's encoding and decoding must be used or can pass empty objects（like nn.nn.Module）

from libmtl.

Actually, I don't understand your question and I hope you can clarify it or give some examples. In fact, encoder_class here must be a class, not an instantiated object, but decoders is the opposite.

from libmtl.

Yes,I didn't express it very clearly
I wondered if it was possible not to use Encoder and decoders class
Because not all models have to be encoded and decoded

from libmtl.

encoder_class and decoders are used to define the network architecture of the model and they are necessary.

The meanings of encoder_class and decoders may be more general. For example, in the hard-parameter sharing pattern, encoder_class represents the task-sharing parameters (usually the feature extractor in cv problems), and decoders denotes the task-specific parameters (usually the output modules).

from libmtl.

In train.py, .next() can't run

I got a same error! -> "AttributeError: 'dict_keyiterator' object has no attribute 'next'"

This method has been deleted in python 3?
Why not use next(iter)?

or what should I do to fix it?

from libmtl.