This the repository of our Heatmap-Guided Balanced Deep Convolution Networks for Family Classification in the Wild.
Requirements :
- PyTorch GPU https://pytorch.org/
- Tensorflow GPU https://www.tensorflow.org/install
- FIW Dataset a. : from https://competitions.codalab.org/competitions/20196#participate-get_data (you may need to register) b. original data from https://web.northeastern.edu/smilelab/fiw/download.html
This repository holds :
- The image normalizer from : https://github.com/deckyal/FADeNN
- The facial landmark localiser from : https://github.com/deckyal/RT
Usage :
Replicate the 2nd challenge of RFIW (https://web.northeastern.edu/smilelab/RFIW2019/)
Preparations
- Put the FIW data on images/ (ex : /DFC/images/FIDs/F0001/MID1)
- Put the test_no_labels.list on the main folder :DFC/
- Run the LandmarkingHeatmap to get the corresponding denoised image and the facial heatmaps
python Reproduce.py
The corresponding CSV will be on the ./models
Replicate the 5 cross validation test of Family classification in the wild (FIW) https://web.northeastern.edu/smilelab/fiw/benchmarks.html
Preparations
- Put the FIWNews data on images/ (ex : /DFC/images/FIDsNew/F0001/MID1)
- Download the five validation split from : https://web.northeastern.edu/smilelab/fiw/download.html and put on the /DFC/cl-info/ folder
- Run the LandmarkingHeatmap to get the corresponding denoised image and the facial heatmaps
- Change testFold in line 124 of Validate pre, ex testFold = 0 signifies the fold 1 test (0~4, fold 1 to 5)
python ReproduceFIW.py
The corresponding CSV will be on the ./models
Citation :
Heatmap-Guided Balanced Deep Convolution Networks for Family Classification in the Wild. [In Recognizing Families In the Wild (RFIW) workshop in conjunction with FG 2019, 2019 May 14, Lille, France]