Commit d5f4c92c authored by Tiago de Freitas Pereira's avatar Tiago de Freitas Pereira
Browse files

Updated chap 5

parent c079de5e
Pipeline #25777 passed with stages
in 11 minutes and 53 seconds
......@@ -5,5 +5,61 @@
Domain Specific Units
======================
This section contains instructions on how to reproduce the experiments from Chapter 5 **Domain Specific Units**.
Exceptionally, these instructions will cover only the **Thermal** database.
This would avoid this section to be extremely large.
However, the same set of instructions applies to ALL heterogeneous face databases.
To see all the available databases, check::
$ resources.py --types database
List of registered databases:
- bob.bio.htface 1.0.0 @ <experiment>/bob.bio.htface:
+ casia-nir-vis-2 --> bob.bio.htface.configs.databases.casia_nir_vis: database
+ cuhk-cufs --> bob.bio.htface.configs.databases.cuhk_cufs: database
+ cuhk-cufsf --> bob.bio.htface.configs.databases.cuhk_cufsf: database
+ eprip --> bob.bio.htface.configs.databases.eprip: database
+ fargo --> bob.bio.htface.configs.databases.fargo: database
+ fargo_depth --> bob.bio.htface.configs.databases.fargo_depth: database
+ ldhf --> bob.bio.htface.configs.databases.ldhf: database
+ nivl --> bob.bio.htface.configs.databases.nivl: database
+ pola_thermal --> bob.bio.htface.configs.databases.pola_thermal: database
+ thermal --> bob.bio.htface.configs.databases.thermal: database
Thermal Experiments
===================
The sequence of experiments in this subsection generates the necessary data that creates Figures 5.16, 5.17 and Table 5.10.
This covers the training using Siamese/Triplet Networks, using Incep. Res. v2 and Incep. Res. v1 as DCNN basis and adaptation of :math:`\beta + W` and :math:`\beta`.
Inception Resnet v2
-------------------
The code below generates the cropped faces that are using to train the DSU for all the cases::
$ bob bio htface htface_baseline htface_idiap_msceleb_inception_v2_centerloss_gray thermal -- preprocess-training-data # generating prior
Siamese training adapting :math:`\beta + W`
-------------------------------------------
The code below trains the DSUs for: :math:`\theta_{[1-1](\beta + W)}`, :math:`\theta_{[1-2](\beta + W)}`, :math:`\theta_{[1-4](\beta + W)}`, :math:`\theta_{[1-5](\beta + W)}` and :math:`\theta_{[1-6](\beta + W)}`::
$ bob bio htface htface_train_dsu siamese_inceptionv2_first_layer_nonshared_batch_norm thermal # Training DSU 1-1
$ bob bio htface htface_train_dsu siamese_inceptionv2_adapt_1_2_nonshared_batch_norm thermal # Training DSU 1-2
$ bob bio htface htface_train_dsu siamese_inceptionv2_adapt_1_4_nonshared_batch_norm thermal # Training DSU 1-4
$ bob bio htface htface_train_dsu siamese_inceptionv2_adapt_1_5_nonshared_batch_norm thermal # Training DSU 1-5
$ bob bio htface htface_train_dsu siamese_inceptionv2_adapt_1_6_nonshared_batch_norm thermal # Training DSU 1-6
With all the DSUs trained, the corresponding experiments are generated via the following bash commands::
$ bob bio htface htface_baseline siamese_inceptionv2_first_layer_nonshared_batch_norm thermal -vv
$ bob bio htface htface_baseline siamese_inceptionv2_adapt_1_2_nonshared_batch_norm thermal -vv
$ bob bio htface htface_baseline siamese_inceptionv2_adapt_1_4_nonshared_batch_norm thermal -vv
$ bob bio htface htface_baseline siamese_inceptionv2_adapt_1_5_nonshared_batch_norm thermal -vv
$ bob bio htface htface_baseline siamese_inceptionv2_adapt_1_6_nonshared_batch_norm thermal -vv
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment