It appears that the colouring strategy used in Middlebury, Flying Chairs and MPI Sintel all use the same strategy, and it's actually not HSV. The approach was originally proposed as part of the Middlebury dataset based on ideas from http://www.quadibloc.com/other/colint.htm
This is an HSV circle
This is the colouring proposed:
Note how in the Middlebury strategy the colour lightness decreases and saturation increases as the flow magnitude increases unlike the HSV approach.
The original code to generate this colouring was provided with the Middlebury dataset:
It appears that the colouring strategy used in Middlebury, Flying Chairs and MPI Sintel all use the same strategy, and it's actually not HSV. The approach was originally proposed as part of the Middlebury dataset based on ideas from http://www.quadibloc.com/other/colint.htm
This is an HSV circle![image](https://user-images.githubusercontent.com/966205/58360906-29495c80-7e83-11e9-88aa-faa8989d1536.png)
This is the colouring proposed:![image](https://user-images.githubusercontent.com/966205/58360877-00c16280-7e83-11e9-9e71-71e5742e790a.png)
Note how in the Middlebury strategy the colour lightness decreases and saturation increases as the flow magnitude increases unlike the HSV approach.
The original code to generate this colouring was provided with the Middlebury dataset:
Deqing Sun also ported it to Matlab
A few people have ported it to Python:
More references: