Intrinsic Image Decomposition from Image Sequence

Python implementation of Y. Weiss, Deriving intrinsic images from image sequences. In Proc. Int. Conf. on Computer Vision (ICCV), volume 2, pages 68-75, 2001

requirement

Implemented with libraries of version in requirement.txt

pip install -r requirement.txt

usage

reproduce example similar to figure 4 in the paper

python demo.py

if you have a directory(folder) containing image sequence:

python demo.py --dir [DIR_PATH]

repalce DIR_PATH with yours.

For example, below is the part of given sequences from MIT Intrinsic Images dataset:

At the below, left is the ground truth and right is derived by algorithm:

And, here is the illumination calculated using derived reflectance:

if you have folder structure like:

folder0

• folder1
  • img0
  • img1
  • ...
• folder2
  • img0
  • img1
  • ...
• ...

you can use save.py as below:

python save.py --input-dir [INPUT_DIR_PATH] --output-dir [OUTPUT_DIR_PATH]

reference

• Y. Weiss, Deriving intrinsic images from image sequences. In Proc. Int. Conf. on Computer Vision (ICCV), volume 2, pages 68-75, 2001
• original matlab code for demo by Y. Weiss (download will start if you click)
• Roger Grosse, Micah K. Johnson, Edward H. Adelson, and William T. Freeman, Ground truth dataset and baseline evaluations for intrinsic image algorithms, in Proceedings of the International Conference on Computer Vision (ICCV), 2009