deepdraw issueshttps://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues2023-04-03T07:53:18Zhttps://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues/27Documentation not linking correctly2023-04-03T07:53:18ZAndré AnjosDocumentation not linking correctlyThe badges of documentation of this package are not pointing to the correct links. Furthermore, I notice this package still has a `master` branch instead of a `main` branch. Shall we change the main branch name and fix the links approp...The badges of documentation of this package are not pointing to the correct links. Furthermore, I notice this package still has a `master` branch instead of a `main` branch. Shall we change the main branch name and fix the links appropriately?https://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues/26Lots of new LFS files in the repository without corresponding instructions2023-04-03T09:24:39ZAndré AnjosLots of new LFS files in the repository without corresponding instructions@dcarron: I see many of our JSON and PNG files (of masks) are now LFS files.
Could you please clarify why these were made LFS? I see, e.g., 5kb files which are now LFS files.
What is the procedure to clone them once the repo is cloned...@dcarron: I see many of our JSON and PNG files (of masks) are now LFS files.
Could you please clarify why these were made LFS? I see, e.g., 5kb files which are now LFS files.
What is the procedure to clone them once the repo is cloned? Unfortunately, this is not clear in the documentation. Some of the students are struggling with new errors.Daniel CARRONDaniel CARRONhttps://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues/24Support for 3D/2.5D images2023-02-15T13:48:50ZAndré AnjosSupport for 3D/2.5D imagesIt was raised during a dedicated meeting, if this package could not support 3D segmentation. The discussed points where:
* Probably models are very different
* Probably the way to structure data for this task is not similar to this par...It was raised during a dedicated meeting, if this package could not support 3D segmentation. The discussed points where:
* Probably models are very different
* Probably the way to structure data for this task is not similar to this particular case
These points push us to implement yet another package to support this. We keep this here, for the time being, in the absence of a better place to report it.Oscar JIMENEZOscar JIMENEZhttps://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues/23Add support for pytorch-segmentation-models2023-02-15T13:43:04ZAndré AnjosAdd support for pytorch-segmentation-modelsWe should consider adding support for the [pytorch-segmentation-models library](https://github.com/qubvel/segmentation_models.pytorch) in this package. It has a bunch of implemented models we can re-use in our benchmarks.We should consider adding support for the [pytorch-segmentation-models library](https://github.com/qubvel/segmentation_models.pytorch) in this package. It has a bunch of implemented models we can re-use in our benchmarks.Maxime DELITROZMaxime DELITROZhttps://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues/22Use pytorch-lightning to replace logging/monitoring and training/evaluation l...2023-02-15T13:45:11ZAndré AnjosUse pytorch-lightning to replace logging/monitoring and training/evaluation loopsClosely related to #21, we should investigate if the use of [pytorch-lightning](https://www.pytorchlightning.ai) is not going to allow us to drop large parts of the code base presented in this package and biosignal/software/ptbench>.
In...Closely related to #21, we should investigate if the use of [pytorch-lightning](https://www.pytorchlightning.ai) is not going to allow us to drop large parts of the code base presented in this package and biosignal/software/ptbench>.
In particular, the following features seem relevant:
* Training and evaluation loops: https://pytorch-lightning.readthedocs.io/en/latest/model/train_model_basic.html
* Logging to [tensorboard](https://www.tensorflow.org/tensorboard) compatible streams: https://pytorch-lightning.readthedocs.io/en/latest/visualize/logging_basic.html
* Early stopping: https://www.youtube.com/watch?v=vfB5Ax6ekHo
* Training on GPUs: https://www.youtube.com/watch?v=nsyVXthGRHoAdopt new tooling (phase 3)Maxime DELITROZMaxime DELITROZhttps://gitlab.idiap.ch/biosignal/software/deepdraw/-/issues/21Move “common” part to new project biosignal/software/basetorch>2023-02-15T13:40:48ZAndré AnjosMove “common” part to new project biosignal/software/basetorch>The key idea is to apply [DRY](https://en.wikipedia.org/wiki/Don't_repeat_yourself) to software packages that use PyTorch inside the biosignal/software group. Example packages are:
* biosignal/software/ptbench>
* biosignal/software/dee...The key idea is to apply [DRY](https://en.wikipedia.org/wiki/Don't_repeat_yourself) to software packages that use PyTorch inside the biosignal/software group. Example packages are:
* biosignal/software/ptbench>
* biosignal/software/deepdraw>
There are several elements that can be shared between these two packages. Most notably:
* Resource monitoring and logging (GPU/CPU utilisation, log files)
* Training and evaluation loops
* Image loading and transformations when applicable
* Base data loading for composed datasets (everything inside the `data` submodule)
* Common analysis code
One key point to keep track relates to using [Pytorch-lightning](https://www.pytorchlightning.ai) as base software to implement monitoring, logging and training/evaluation loops. One must decide if we do this now, or later. If done later, we risk to redo and remove common API work to be done now. If done now, at this stage, then one must make sure to implement this through various MRs, each addressing only one aspect so we can keep track of changes.Building a common base (phase 2)Daniel CARRONDaniel CARRON