Usage of MFFV_code

code for reproducing MFFV experiments

1. Processing of the experiments

2. Structure of the code

3. Reproducing The Experimental Results

Step1: Read the paper, and the structure of the code (above), understand the functions of each script.

In the following commands, the folder bob.paper.tifs.mffv is your local folder where you have cloned this repository.

• activate your conda bob environment.
• The dataset path is: /idiap/resource/database/MFFV-N/data

Step2: Feature extraction:

Use the following commands for feature extraction

cdbob.paper.tifs.mffv/FEpythonfeature_extraction.py-db/path/of/the/MFFV/dataset/-fe/path/of/the/feature/you/want/to/save/

'feature_extraction.py' can also be run on the SGE-grid. If using grid, it should be distributed over 5760 jobs.

Step3: FVIA experiments (Section V-D). Details are as follows:

Use the following commands to run the whole FVIA experiments (Table VII-X). (When the first command finishes you can run the next command.)

cdbob.paper.tifs.mffv/FVIApythonfvia.py-db/path/of/the/MFFV/dataset/-fe/path/of/the/feature/you/had/saved/-pt/path/of/the/protocols(PTfolder)/-rs/path/of/the/intermediate-results/you/want/to/save/

once the above experiment finish, conduct the score-level fusion experiment:

pythonfusion_fvia.py-rs/path/of/the/intermediate-results/you/have/saved/-fs/path/of/the/intermediate-results/you/want/to/save/# recommendation: set these two paths to be the same

'fvia.py' can also be run on the SGE-grid. If using grid, it should be distributed over 81 jobs.

• cd to path: /path/of/the/intermediate-results/you/have/save/, then use the following commands for each result-file (example: 'dev_balance_c1_m6_th120.csv') to get the final results/metrics

bobbiometrics-d2-cmin-hterdev_balance_c1_m6_th120.csv

Step4: FVAB experiments (Section V-E). Details are as follows:

Use the following commands to run the whole FVAB experiments (Table XI). (When the first command finishes you can run the next command.)

cdbob.paper.tifs.mffv/FVABpythonfvab.py-db-db/path/of/the/MFFV/dataset/-fe/path/of/the/feature/you/had/saved/-pt/path/of/the/protocols(PTfolder)/-rs/path/of/the/intermediate-results/you/want/to/save/

once the above experiment finish, conduct the score-level fusion experiment:

pythonfusion_fva.py-rs/path/of/the/intermediate-results/you/have/saved/-fs/path/of/the/intermediate-results/you/want/to/save/# recommendation: set these two paths to be the same

• cd to path: /path/of/the/intermediate-results/you/have/saved/, then use the following commands for each result-file (example: dev_xxx.csv, and test_xxx.csv ) to get the final results/metrics

bobbiometric-e-d2-ceerdev_xxx.csvtest_xxx.csvbobbiometric-e-d2-cmin-hterdev_xxx.csvtest_xxx.csvbobbiometric-e-d2-cfar-f0.01dev_xxx.csvtest_xxx.csvbobbiometric-e-d2-cfar-f0.001dev_xxx.csvtest_xxx.csv

4. Dataset

To get the MFFV-N dataset, please sign a license agreement.

5. Contact

For further questions/discussions/cooperations, contact Junduan Huang, Email: runrunjun@163.com