Anaglyph HALD Generator (Anahald) - Auto-choice of a LUT for stereopair

Required are Python interpreter and the following libraries: torch, torchvision, pandas.
The need for additional libraries complicates use-cases where the libraries are installed under virtual environments.
Having said the above, most Python installations should support command-line usage scenario - when the user first opens the console window with the required environment, then types commands into it.
On contrary, drag-and-drop kind of usage requires to start python under the hood, so that there's no chance to request a particular environment.

The author succeeded to use drag-and-drop only with WinPython on Windows, while command line worked under WinPython and Anaconda on Windows, and under pre-installed Python 3.8 under WSL-Linux.

If you are on Linux, Python is most probably preinstalled; otherwise follow the installation procedure of your Linux distribution.
If you are on Windows, the author's recommendation is to install WinPython - it should handle both command-line and drag-and-drop usage approaches.

Do install the 3 additional libraries - torch, torchvision, pandas - by issuing the three "pip install LIBRARY-NAME" commands at the OS prompt (i.e. CMD on Windows, BASH on Unix):

• On Windows it should be CMD terminal associatd with your Python installation - like "WinPython Command Prompt" for WinPython, "Anaconda Termimal" for Anaconda-Python.
  
• On Linux it should be a regular BASH prompt.
  

The last step is to "help" Anahald auto-choice in finding the appropriate Python interpreter. This is achieved by one-line scripts Code/Choice/python.bat and Code/Choice/python.sh that invoke Python interpreter by absolute path while passing it all command-line arguments.
The distribution includes 3 examples of such files:

• Code/Choice/python.bat.EXAMPLE - for use with DOS-CMD scripts on Windows
  
• Code/Choice/python.sh.EXAMPLE-WINDOWS - for use with BASH scripts on Windows
  
• Code/Choice/python.sh.EXAMPLE-Linux - for use with BASH scripts on Linux
  

Copy the relevant example into script named Code/Choice/python.bat or Code/Choice/python.sh accordingly, and change the path to match your system.
Note that Linux example doesn't include absolute path; it relies on the system PATH instead; it's fine on Linux where it's unlikely to have multiple installations of Python

See Obtaining ImageMagick for installation instructions.
After installing ImageMagick, find the path of its main utility - "magick[.exe]" for newer versions or "convert[.exe]" for older versions.
You need to define environment variable IMAGEMAGICK_CONVERT_OR_MAGICK to contain the above path.
If ImageMagick is included in the system PATH variable, IMAGEMAGICK_CONVERT_OR_MAGICK can contain just "magick[.exe]" or "convert[.exe]"; otherwise absolute path is required. The former option is typical of Linux systems, while the latter is more suitable for Windows.

Anahald auto-choice expects a hardcoded folder hierarchy to locate its code- and data files:

  Directory arrangement:
----------------------
      +---SAMPLE_HALDS                      
      |   |   hald__ahg_oleg_cp__16.TIF     
      |   |   hald__ahg_oleg_ec__16.TIF     
      |   |   hald__ahg_oleg_gp__16.TIF     
      |   |   hald__ahg_oleg_sf__16.TIF     
      |   |                                 
      |   +---ADD                           
      |           hald__ahg_oleg_mc__16.TIF 
      |           hald__ahg_oleg_xc__16.TIF 
      |                                     
      +---Code                           
      |   |                                 
      |   +---Choice         
      |   |   (Python code files)
      |   |   choose_hald_for_image.bat
      |   |   choose_hald_for_image.sh
      |   |   python.bat
      |   |   python.sh
      |   |                                 
      |   +---View         
      |   |   (Python code files)
      |
      +---CHOICE_MODELS
      |   |   anahald_label_codes__96d5__20250814-231159.csv
      |   |   anahald_model_params__96d5__20250814-231159.pth
      |

• Anahald choice code is a part of Anahald distribution; follow the instructions in Anahald source code installation guide if you didn't do it yet.
  
  • Note that the link here points at the specific subsection dealing with obtaining source code. If you plan to use pre-made LUTs and don't intend to generate new ones, you don't need to neither install Tcl, nor to verify that Anahald Tcl code functions.
    
• Download Anahald premade LUTs - sample HALDs here and place them under directory ..\..\SAMPLE_HALDS\ relative to location of 'choose_hald_for_image.bat'.
  
• Download Anahald choice model files here and place them under directory ..\..\CHOICE_MODELS\ relative to location of 'choose_hald_for_image.bat'.
  

choose_hald_for_image.bat INPUT-SBS-IMAGE-GLOB-1 … INPUT-SBS-IMAGE-GLOB-n [/OUTPUT-DIRECTORY/]

• INPUT-SBS-IMAGE-GLOB - either path of SBS image file, or glob pattern matching one or more such files; e.g. the script obtains either list of input files, or one or more glob patterns matching those
  
• OUTPUT-DIRECTORY - optional path of the output directory for resulting anaglyphs; if not given, defaults to "ANA\" subdirectory under the location of input file(s)
  

choose_hald_for_image.sh INPUT-SBS-IMAGE-GLOB-1 … INPUT-SBS-IMAGE-GLOB-n [/OUTPUT-DIRECTORY/]

• parameters are the same as for Windows CMD; only path syntax differ
  

Example (BASH on Windows):

/c/Oleg/GitWork/AnaHald/Code/Choice/choose_hald_for_image.sh  /G/Photo_Archive/Photos_5/Israel_2016/Stereo/241216__Akko3D/SBS/*.TIF  /y/TMP

Drag-and-drop invocation differs in having no provision to supply command arguments other than list of input files. For this reason output directory will always be "ANA\" subdirectory under the location of input file(s), and glob patterns aren't supported.
For convenience, the author's recommendation is to place shortcut to choose_hald_for_image.bat on the desktop.
After one or more input images are dropped onto the shortcut, a new DOS-CMD window opens where operation log is printed. If anything goes wrong, examine error and warning messages in this window.
When the script finishes, "Press Enter to close…" prompt appears in the bottom of the above log window. Do not press Enter if you want to read the log.

1. A collection of SBS stereopairs
   

2. A CSV file that maps between the above stereopair filenames and sample HALD LUTs chosen for them. Example of a CSV file:
   

   $ head -6 sbs_to_hald.csv
   SbsFileName,HaldId
   DSC00022-00034,ahg_oleg_cp
   DSC00033,ahg_oleg_gp
   DSC00035,ahg_oleg_gp
   DSC00037,ahg_oleg_gp
   DSC00040-00564,ahg_oleg_sf
   

Here it's assumed that you have Anahald installed (which includes auto-choice) - e.g. Pre-requisites and installation section is completed.

• Arrange Python environment, preferably with CUDA videocard processor(s), install libraries torch, torchvision, and pandas.
  
  • in order to have CUDA available, you may need Python interpreter different from the one used for auto-choice invocations
    
• In code file "choose_hald_base.py" change the target accuracy - training stops and stores the results when accuracy indicated by "SUFFICIENT_ACCURACY_TO_STOP" variable is achieved.
  

• As shown in the following example, load the code into the above Python shell and start training. Note the addition of Python-code path to sys.path; do provide the path that matches your Anahald installation.
  

  import sys;  sys.path.append('C:\\Oleg\\Gitwork\\Anahald\\Code\\Choice')
  from choose_hald_resnet import *
  
  TrainAnahaldResnet('sbs_to_hald.csv', 'ALL_SBS_1080', num_epochs=100)
  

  Comments for the example:
  

  • directory added to sys.path is the one where "choose_hald_resnet.py" is located
    
  • 'sbs_to_hald.csv' is the path of CSV file mapping stereopair filenames to HALD identifiers
    
    • the HALD identifier is not its filename; for example sample HALD file "hald__ahg_oleg_cp__16.TIF" has identifier "ahg_oleg_cp"
      
  • 'ALL_SBS_1080' is the path of a directory with input stereopair images
    

Two files constitute training output; the <TIMESTAMP> fields in their names must be equal:

1. "anahald_model_params__<TIMESTAMP>.pth" - the trained model parameters
   
2. "anahald_label_codes__<TIMESTAMP>.csv - maps internal label codes to HALD identifiers
   

Anahald training reports the paths of output files; watch for messages like:

-I- Saved model in 'MODELS\anahald_model_params__20250817-221920.pth'
-I- Success saving dictionary in 'MODELS\anahald_label_codes__20250817-221920.csv

Copy the two files into "CHOICE_MODELS" directory (see Code- and Data files).
Update the model filename in "choose_hald_for_image.py" - change the assignment of variable "MODEL_FILENAME", e.g.:

MODEL_FILENAME = "anahald_model_params__20250817-221920.pth"

Now everything is set to run auto-choice with the new model.