Merge Gui into main

[davidpagnon]

Merge Gui branch (@AYLARDJ) into main

[davidpagnon]

Thanks Yacine for this great work!
This is quite close to a version we can publish :)
I edited a few things, and I have a few other suggestions:

What I did:

• Run as CLI: Launch the GUI by running the command Pose2Sim from anywhere on your computer (I just added it as a script in pyproject.toml, and edited the imports a little). The exe file can also be found in C:\Users\USERNAME\miniconda3\Scripts.
• Missing dependencies: I added customtkinter and requests in the dependencies. I did not add tomlkit though, as I'd rather keep working with toml, so I replaced all the occurrences of tomlkit.
• Link to files rather than copying them: Instead of copying the videos and various files, which could very quickly add up storage space, I created a symbolic link to the original files.
• Imported config from library: Instead of hardcoding the whole configuration to the python file, I loaded it from Pose2Sim or Sports2D (with Path(Pose2Sim.__file__).parent / 'Demo_SinglePerson' / 'Config.toml'). There are several occurrences where the configuration options are hardcoded: eventually, it would be nice to get them directly from the default calibration file, since the configuration file will probably evolve with time and maintaining the GUI would become complicated.
• Enabled other pose estimation models
• Merged HunMin's intro screen, slightly changed the timings. Could not find a way to fix the error: "invalid command name "2568299333440update" while executing "2568299333440update" ("after" script)"
• Removed GUI.py
• Replaced a large part of the os occurrences with Path from the pathlib library, which is more modern. I did not do app.py, blur.py, utils.py. Evertually, I'd like to replace them all, including in the core package. Also removed some imports that were not used
• Used subprocess instead of os.system + start, in order to run cross-platform

[davidpagnon]

My suggestions, now. I did not reread it, so feel free to tell me if anything is unclear!

My suggestions:

General:

• Some Config parameters from the newer Pose2Sim and Sports2D version are missing. For Sports2D, it makes the analysis fail.
• Since we are picking the files by hand, I do not think it is necessary for the user to specify file extensions
• Would be nice to have buttons a bit more descriptive than "Proceed with...", and a bit fewer pop-up windows. I'm giving you an example in the Calibration section
• It would also be nice to add a green "Next" button that sends you to the next stage. Even better, there could be a last button that says "Run now" if the user wants to only run calibration or any single stage
• Some occurrences of the os library still need to be changed to their pathlib equivalent (not important): app.py, blur.py, utils.py

Tutorial:

• tutorial_preview.png not available
• Not sure we need both the buttons "skip tutorial" and "complete tutorial"?

Calibration:

• Cannot edit calibration checkerboard once it has been confirmed
• When Convert is selected: should not have a number of cameras, and No checkerboard preview
• Would be good to be able to select multiple files (in the case of images, for example)
• Instead of the button "Proceed with calibration" which is not very clear, and the popup window that says "Please select the checkerboard etc", you could only have more buttons, for example Intrinsics: "Select checkerboard images/videos" and Extrinsics "Select scene images" and "Define scene coordinates"
• Scene point selection: I think it would be good to either have a popup window with (X, Y, Z) fields right after each click, or one single window with one line per point, with three (X, Y, Z) columns

Prepare videos:

• I'm not entirely sure why we need two modes. Here is what I suggest, but feel free to tell me if I missed something:
  • Simple mode is only useful to extract checkerboard images, right? If so, it could be an option in the previous Calibration tab
  • Advanced mode is great (although I have not fully tested it), but in most cases, the user won't need it. What about renaming the tab "Edit videos (optional)
• I think the sections should be more obvious:: File operations, Drawing, Cropping, Rotation, Face Detection. There problem right now is that the shubsections (shape list, for example) are at the same level as the main ones.
• Do we need the shapes to have an outline color? I feel like they could just be dashed lines.
• Do we need the buttons "Enable ..." (video cropping, rotation, etc)?
• You could also use the new face_bluring.py tool for automatically bluring faces (I would actually put it at the very top, as it might be the most used tool in the future)
• the pixelation effect could be stronger
• I got a few bugs, I hope it is not me, but it is entirely possible that I messed up with your files:
  • Prepare video -> Only checkerboard images ->
    File "D:\softs\github_david\pose2sim\GUI\tabs\prepare_video_tab.py", line 340, in confirm_checkerboard_only
    for widget in self.frame.winfo_descendants():
    AttributeError: 'CTkFrame' object has no attribute 'winfo_descendants'
  • Upon lauching video editor:
    File "D:\softs\github_david\pose2sim\GUI\tabs\prepare_video_tab.py", line 340, in confirm_checkerboard_only
    for widget in self.frame.winfo_descendants():
    AttributeError: 'CTkFrame' object has no attribute 'winfo_descendants'
    Error initializing RTMPose:

Data visualization:

• TRC plot: In case it can help, have a look at this:
  • The scale is adjusted at each time frame. It would be better to scale it from the maximum X,Y,Z values across all frames
  • The scales are not equal on the X,Y,Z axes, which makes the bodies deform at times
  • Axes are not in the right order. 3D coordinates (X forward, Y left, Z up) becomes in the TRC frame of reference (X left, Y up, Z forward), and in Blender (X right, Y forward, Z up)
• A controlable timeline would be great

Not sure how to go about it, but it would be nice to integrate Mstudio, too

[davidpagnon]

I made a first attempt at making a one-click install. I haven't fully tested it yet, but basically, here is what I want:

• To create the installer, run this command line:
  pyinstaller --onefile --name Pose2SimInstaller --windowed --icon=GUI/assets/pose2sim_logo.ico pose2sim_installer.py.
  This creates a Pose2SimInstaller.exe file.
• Double-click this file to launch an installer GUI (or not if you add the argument --console). This will automatically install conda, python, and the pose2sim libraries.
• Once installed, a Pose2Sim.exe script is created on the desktop, that launches the actual GUI from Yacine.

However:

• It needs to be fully tested (I haven't even tried to finish the installation)
• It would be even greater if we updated the .github/workflows/publish_on_release.yml script so that it automatically creates a new installer upon each release, saves it in the dist folder, that users could directly download

[davidpagnon]

Calibration:

• added an option to calibrate with a vertical checkerboard
• save images with reprojected points

Fixed edge cases:

• handles trimmed_mean function failure (when computing height and scaling)
• fixed frame range issues in pose estimation, triangulation, filtering, marker augmentation
• handled cases when no videos found for synchronization GUI and fps determination