Error while testing and training with 'LmdbDataset' object has no attribute 'max_len'
Opened this issue · 20 comments
when I checked the code there are some required fields but did not assign any values. I also faced max_len problem.
@adeline-cs
Thank you for code released. Your work looks very exciting so many researchers want to try. could you please fix the bug in this repo. so that we can try your network and we will let you know if the problem exists. now we can't train your network because not clean code.
@lerndeep @bharatsubedi
Thank you. I will revise as soon as possible, and update these two days.
@adeline-cs waiting for release clean code. are you planning to publish or not?
@adeline-cs I checked your current update still not solved max_len problem. if you check inside SGTR.py you are not assigning any parameters for different modules. I was thinking you will update code that will be run, but you are not fixing the problem exists till.
@adeline-cs I am also waiting for code without error. could you please let me know are going to publish it or not?
geo_map list is empty.
@adeline-cs Thank you very much, I read your paper it is very interesting and looks faster than other SOTA methods so waiting for bug-free code.
@WongVi @lerndeep We will update the github about the problem of some detail setting for you, and explan in readme file. Thank you all.
Thanks @adeline-cs Even iam waiting for the bug free code.
@WongVi For the problem of "geo_map", which is relevant to our adaptive training strategy. But this part of the adaptive training code has not been released yet. We will provide simple and usable code as soon as possible and release the full adaptive training code at a later stage.
For the problem of "geo_map", which is relevant to our adaptive training strategy. But this part of the adaptive training code has not been released yet. We will provide simple and usable code as soon as possible and release the full adaptive training code at a later stage.
waiting
Was this ever resolved?
Nothing in the entire repository works tbh. Idk why it would take so much time if they have already ran so much extensive testing while publishing the paper.
None of the code actually works...
The S-GTR model gives several import errors and tries to import files that seem to not even exist
GTR-CRNN and GTR-TRBA etc. plugin models do not even have, use, or load a GTR model from the looks of it. I can't find any import or text string in the code that says graph or gtr (case insensitive search) anywhere. They also have several import errors on top.
All of this makes me think that either the results are not actually as good.. or they have some pretty badly maintained code that they need to fix. Seems no point in releasing several pretrained models and their code if you cannot even run a single one of them. Perhaps just release a propper working S-GTR baseline first...
Still makes no sense why it would take so long to release a working version of S-GTR baseline that we can train on our own...
Plenty of import errors referencing completely non existent packages/modules everywhere.
Probably going to have to run some other models for our purposes at this point because no idea how long it might take for an even slightly replicatable release of the code...
They did update the package with a commit recently, so I'm curious whether or not they plan to resolve this.
They did update the package with a commit recently, so I'm curious whether or not they plan to resolve this.
That didn't change anything meaningful. Just some stuff in the readme and such.
In code, they just commented a line and copy pasted the same line below it, making it remain the exact same.
Thanks to @siddagra @macksjeremy. We recently checked the import error of the plugin code and solved the problem. Due to doing other projects, we plan to release some sample packages later for S-GTR. Please wait for a while.
@adeline-cs I check your log before the long time you just told that you will open the clean code but not been fixed yet. I don't know if you want to open it or not if you want to open please open the first S-GTR clean code.
Thank you for your suggestion and your attention@WongVi. We will reorganize the code structure, and give clean code for S-GTR and plugin models in separated branches. The new code can solve the above problems. Please wait patiently.
Excuse me, is this project updated now, can the code run normally?