prepCV 2.1.0

Define preprocessing pipelines using nested callable functions and parameter grids for them.

Auto-Preprocessing for Computer Vision

This Python module helps you automate the selection of the best preprocessing pipeline for your computer vision tasks. It uses a combination of parameter grid search, interactive visualization, and caching to efficiently find and apply optimal preprocessing steps to your images.

Features

• Define Multiple Pipelines: Create and store multiple preprocessing pipelines, each with different sequences of cv2 functions and a range of parameter values.
• Grid Search: The module performs an exhaustive grid search across all defined pipelines and parameter combinations.
• Interactive Visualization: A user-friendly image selector allows you to visually compare the results of different preprocessing steps and choose the best one.
• Caching: The module caches the results of the search process, so repeated runs with the same pipelines and image will be much faster.
• Preprocessor Construction: Once the best pipeline is selected, the module automatically constructs a Preprocessor object that you can use to apply the optimal preprocessing to new images.

Installation

1. Git Clone: Clone this repository into your working directory:

   git clone https://github.com/Kalmy8/auto_preprocessing
   

2. Requirements: Make sure you have the required libraries installed:

   pip install -r requirements.txt
   

Alternatively:

1. Install as package using pip:

   pip install prepCV
   

Usage

1. Define Pipelines: Create PipelineDescription objects to define your preprocessing pipelines.

   import cv2
   from prepCV import PipelineDescription, PipelineManager
   
   pipeline1 = PipelineDescription({
       cv2.cvtColor: {'code': [cv2.COLOR_BGR2GRAY]},
       cv2.adaptiveThreshold: {
           'maxValue': [255],
           'adaptiveMethod': [cv2.ADAPTIVE_THRESH_MEAN_C], 
           # ... other parameters ...
       },
       # ... other cv2 functions ...
   })
   
   pipeline2 = PipelineDescription({
       # ... define another pipeline ...
   })
   

2. Add Pipelines to Manager:

   pipeline_manage = PipelineManager()
   pipeline_manage.add_pipeline(pipeline1)
   pipeline_manage.add_pipeline(pipeline2) 
   

3. Run Search:

   # Load your image
   image = cv2.imread('your_image.jpg')
   
   # Run the search and select the best preprocessor
   pipeline_manage.run_search(image, 'GridSearch') 
   
   # Get the best preprocessor
   best_preprocessor = pipeline_manage.get_best_preprocessor()
   

4. Save Search Results to cache:

   from prepCV import CacheManager
   
   # Restore previous search result 
   pipeline_manage = PipelineManager()
   pipeline_manage.load_from_cache()

   # Degine some new pipeline
      pipeline3 = PipelineDescription({
       # ... define another pipeline ...
   })
   
   # Add pipelines to PipelineManager
   pipeline_manage.add_pipeline(pipeline1)
   pipeline_manage.add_pipeline(pipeline2)
   pipeline_manage.add_pipeline(pipeline3)
   
   # This search will only compare new pipeline3 to previous best seen pipeline
   pipeline_manage.run_search(image, 'GridSearch') 

   # Save results to cache
   pipeline_manage.save_to_cache()

5. Use the Preprocessor:

   processed_image = best_preprocessor.process(new_image)
   # Now you can use the `processed_image` for your tasks.