Modification of the definition of networks

dhlab-epfl / dhSegment

Generic framework for historical document processing

https://dhlab-epfl.github.com/dhSegment

GNU General Public License v3.0

373 stars 116 forks source link

Modification of the definition of networks #18

Open SeguinBe opened 6 years ago

SeguinBe commented 6 years ago

So now the network definition is based on two bases classes, an Encoder and a Decoder.

This simplifies quite a bit of things, and allows more flexibility. Also, it will allows anyone to plug-in its own network definition without having to clone dh_segment.

SeguinBe commented 6 years ago

Uh why close it?

Le lun. 5 nov. 2018 à 09:40, solivr notifications@github.com a écrit :

Closed #18 https://github.com/dhlab-epfl/dhSegment/pull/18.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/dhlab-epfl/dhSegment/pull/18#event-1945592366, or mute the thread https://github.com/notifications/unsubscribe-auth/AGzWkTKqx6tNRmO80Gy6eUx2yGarkEHJks5ur_l9gaJpZM4XZ-ok .

solivr commented 6 years ago

oops sorry I closed doc_refactor branch, but forgot that this one is related to it...

wrznr commented 5 years ago

Is it recommended to use this branch instead of master? It seems to have a lot of fixes and extensions...

rkeiii commented 5 years ago

I just wanted to note I was unable to run demo.py on the master branch or the latest release tag. It would fail with a ValueError. That being said, demo.py ran just fine on this encoder_revamp branch. It's somewhat confusing to a newcomer like me that this isn't noted anywhere.

wrznr commented 5 years ago

A lot of the changes suggested here are actually part of the master branch. :confused:

solivr commented 5 years ago

@wrznr The official branch is the master branch. This one should work as well, but I'm not sure it has been tested and we need to make sure it behaves like master (or better) before merging it. It will hopefully be merged some day but I am not working on this at the moment, and AFAIK no one else is. Sorry for that.

SeguinBe commented 5 years ago

To pursue on @solivr comment, we do not have anybody continuing the development at the moment ( in fact I have not been part of the lab anymore for some time already).

A lot of the commits are actually merged from master though so it might appear that there is much more modifications than there are actually.

SeguinBe commented 5 years ago

We did a bit of merging today, things left to correct.

[x] check the performance between this branch and the old one to verify it is still working properly
[x] update the setup.py packages for the version numbers to be more flexible
[x] update the installation documentation so that pip install after tensorflow installation is the preferred method (reminder: tensorflow can be installed with anaconda or pip, then pip install dh-segment will not install tensorflow)
[ ] update the general documentation
[ ] update the demo, ideally a zip file containing training data and a config file.

solivr commented 5 years ago

I have trained two models for each branch on two different tasks (baseline and page extraction). The optimal post-processing parameters seem to be different for the two branches, but models can achieve similar results.
The package versions are now more flexible in setup.py (in master).
Installation is now updated (pip install followed by conda's tensorflow installation) (in master).
Two examples of data generation, training and inference are now available in exps folder (master).

This branch seems ready to be merged with master.