readicculus / pep_gui Goto Github PK

model files are big and don't want to include in sources.

not sure if seal-tk can be launched with input arguments for dets/image lists but would be useful to be able to launch seal-tk for a given dataset from this GUI.

Feedback/notes/questions from Erin:

• Add to menu image list check
• I need to make settings user settings because people like to change things
• Make tool notify if output not same size as input. Warn output length not same as input and list which ones did not get run.
  • Possibly have a column in manifest for if image lists should be aligned
• Copy input list(?)

Empty lines in image list what will happen?

Develop ability for a run to be contained within a 'session'. A 'session' has state and will allow for:

• Serializable context
  • Make context serializable so that a run can be reproduced
    • Include model/pipeline info
    • Configuration parameters used
    • Input image lists
    • Saving checkpoints/allow to resume a partially completed session/task
• Session has a working directory for all resulting artifacts
• Logging
  • Kwiver/viame logs
  • Make it clear to a user when a single dataset is complete

Would be nice to be able to have model be a parameter and have it be interchangeable within a pipeline to avoid the many errors we've seen caused by copying and modifying pipelines just so we can replace the model block. Less files thrown around is a plus too.

tabs on left.
one tab per dataset.
more room to display progress and user input buttons for each dataset as well as output log than if its all on one big scrolling page.

Test on windows.
Fix issues.
Profit?

• Add menu for resuming job from properties window
• Make so if error properties window closes

In pipeline setup need to do a check to ensure selected datasets and pipelines are compatible.

If for some reason files the GUI is trying to use the user doesn't have permissions for show useful error in gui.

• Allow for user defined detection output location
  • save it as a setting
  • only move file there when a task is completed
• Copy inputs as well (image list, maybe transformation?) to the output location

Allow relative paths for dataset manifest so that a dataset can be shared on multiple machines

Issue with multiple users using at once and changing stuff, thought there wouldn't be much need from people to change stuff but guess there is.

Include full network path, rather than just image name, in outputs from processing jobs

when user selects a dataset manifest, need to give useful errors if format or something is wrong.
Also should check if the folders/files defined in the dataset manfiest actually exist and warn or error if they do not.

• Imports Test
• Dataset Manifest Test
• Pipeline Manifest Test
• Test running kwiver runner in scheduler with event manager.
• Pipeline Config Test
• Config Serialization Test
• Job Meta Serialization Test
• Pipeline Serialization/Compiling Test

instead of having to define imagelists would be useful to be able to instead define a wildcard
path/to/image/directory/*_ir.tif

Backend ability to resume already works, just need to make it so that user can do this through the ui.

Allow user to save their own configurations and to load from a saved configuration during the configuration step.

Improve the dataset selection:

• Add a way to select all visible, maybe a shift select as well
• Add a 'deselect all' button
• Add a clear filter button
• Add a reset button

Document:

• How to use GUIs
• Outputs from running a job
• Dataset manifest file specification
• How to report issues (how to send log, error log, ask for screenshot etc...)