An open API service indexing awesome lists of open source software.

https://github.com/ackness/preprocess_mpiigaze


https://github.com/ackness/preprocess_mpiigaze

Last synced: 3 days ago
JSON representation

Awesome Lists containing this project

README

          

# MPIIGaze Preprocessing

## Info

First you should download dataset from [MPIIGaze](https://www.mpi-inf.mpg.de/departments/computer-vision-and-machine-learning/research/gaze-based-human-computer-interaction/appearance-based-gaze-estimation-in-the-wild).

[Official MPIIGaze Normalize Code](http://datasets.d2.mpi-inf.mpg.de//MPIIGaze/data_normalization_code.zip) were written by matlab code and return `Gray` eye regions images.

These repo is rewritten by python according to the matlab source code and return `RGB` images.

## Usage

```bash
python preprocess_mpii.py -p "path/to/mpii/Original"

for example:
python preprocess_mpii.py -p ./MPIIGaze/Data/Original/
```

You will get a dict which seems like this:

```
{'p02/day07/0001.jpg':
{'right_eye':
{'img': array([[[ 89, 80, 95],
[ 95, 86, 100],
[100, 87, 97],
...,
[ 73, 45, 46],
[ 84, 51, 54],
[ 91, 58, 62]],

[[104, 93, 107],
[109, 96, 110],
[111, 96, 108],
...,
[ 81, 51, 53],
[ 92, 58, 62],
[100, 65, 69]],

[[117, 103, 113],
[120, 105, 116],
[120, 105, 115],
...,
[ 92, 57, 61],
[ 98, 61, 66],
[105, 67, 72]],

...,

[[184, 179, 190],
[182, 178, 187],
[179, 174, 185],
...,
[106, 82, 85],
[113, 90, 93],
[120, 97, 100]],

[[183, 176, 189],
[184, 177, 189],
[185, 177, 190],
...,
[116, 92, 96],
[124, 100, 103],
[130, 107, 110]],

[[187, 178, 193],
[188, 179, 192],
[189, 180, 192],
...,
[129, 107, 109],
[137, 116, 117],
[142, 121, 123]]], dtype=uint8),
'headpose': array([0.25371083, 0.13010662]),
'gaze': array([-0.03550274, 0.08322771])},

...
}
```