CCV23.IITP

This repo for course of color computational photography for bachelors in IITP'23

Course program

Lec 1. Introduction to color
Lec 2. Introduction to pipeline
Lec 3. Practical introduction to color
Lec 4. Debayering
Lec 5. Denoising + Illumination Estimation + Illumination Compensation + Color space transform

Mini course

Lec 1. Datasets in Computer Vison: Importance, Theoretical background and Typical Problems in Applications
Lec 2. ML System Design: Bridging the Gap Between a Vanilla Neural Network and Value for User

Course scoring

Since schedule of this course was unstable, score will be calculated accorging to wollowing rules

$Score = 0.8 \times Exam + 0.2 \times Presense + 0.5 \times Practice$,

$Exam$ - is calculated after oral converstion about two topics from the list;
$Presense$ - is a percent of lectures you have visited;
$Practice$ - is a score for your solution on pipeline reverse engineering.

Homework (Deadline: 19.12)

This year, the reverse engineering of the Canon 600D camera will be a homework assignment. To do this, download the Cube++ dataset and create an algorithm reproducing the pipeline of image processing of the Canon 600D camera. Your solution will be tested on new images from the Canon 600D camera.

Installation and requirements

To work with code it is recommended to use Python 3.9+. The required packages are listed in requirements.txt and can be installed by calling:

pip install -r requirements.txt

raw_prc_pipeline

Module raw_prc_pipeline contains the source code of various methods and functions that can be used for processing raw images, including:

• Parsing metadata (see raw_prc_pipeline/exif_*.py files)
• Demosaicing, white_balancing (Gray World, White Patch, Shades of Gray, Improved White Patch), tone mapping and other methods (see raw_prc_pipeline/pipeline_utils.py)
• Implemtations of demo classes of raw image processing pipeline and image processing pipeline executor (see raw_prc_pipeline/pipeline.py)

data

Directory data conatins example of challenge data:

• PNG file with raw image data (see data/IMG_2304.png) and
• Corresponding JSON file (see data/IMG_2304.json) with necessary metadata including: black_level, white_level, cfa_pattern, color_matrix_* and etc. Metadata was extracted using raw_prc_pipeline module.

demo

Directory demo contains demonstration script for processing PNG raw images with JSON metadata using implemented classes and finctions from raw_prc_pipeline.

To process PNG images with corresponding metadata from data directory call the following command:

python -m demo.process_pngs -p data -ie gw -tm Flash

To see other arguments of the script call python -m demo.process_pngs -h from the root directory.

Verification

Each student who decided to develop his own pipeline will have to send me a code to render jpegs. An input for your pipeline shoud be .PNG and .JSON files same as in data folder, output should be .JPEG