-
Notifications
You must be signed in to change notification settings - Fork 11
File Formats
The main files encode similarity information. The files are supported both in the ranked lists (numeric or string) or matrices (similarity or distance) format. These formats are available for input and output.
Ranked lists can be represented in both numeric or string format.
Each line corresponds to a ranked list and the name of the images are separated by common spaces. The names must match the ones specified in the lists file.
apple1.png apple2.png bird1.png bat1.png bat2.png bird2.png
apple2.png apple1.png bird2.png bird1.png bat1.png bat2.png
bird1.png bird2.png bat2.png apple2.png apple1.png bat1.png
bird2.png bird1.png bat2.png apple1.png apple2.png bat1.png
bat1.png bat2.png apple1.png apple2.png bird2.png bird1.png
bat2.png apple1.png apple2.png bat1.png bird2.png bird1.png
Each line corresponds to a ranked list and the number of the images are separated by common spaces.
0 1 2 4 5 3
1 0 3 2 4 5
2 3 5 1 0 4
3 2 5 0 1 4
4 5 0 1 3 2
5 0 1 4 3 2
Matrices can be represented in two different types: similarity and distance.
The values are separated by common spaces. In such a way that line x and row y provide the distance/similarity of image x to image y.
- NOTE: In this case, float values are supported.
0.0 6.1 1.0 9.6 5.6 4.5
6.1 0.0 4.8 5.1 7.6 3.5
1.0 4.8 0.0 7.2 3.1 7.1
9.6 5.1 7.2 0.0 2.5 9.6
5.6 7.6 3.1 2.5 0.0 1.1
4.5 3.5 7.3 9.6 1.1 0.0
The lists file considers each line as an identifier for a dataset element and it is always required for the execution.
apple1.png
apple2.png
bird1.png
bird2.png
bat1.png
bat2.png
For the effectiveness evaluation, each image is associated to a class, information that is defined in the classes file using the expression: image:class. The names must match the ones specified in the lists file.
-
NOTE: This file is only required when
EFFECTIVENESS_EVALUATION
is enabled.
apple1.png:apple
apple2.png:apple
bird1.png:bird
bird2.png:bird
bat1.png:bat
bat2.png:bat