VisualizingNDF

The repository contains pre-processed images, pre-trained models and code for visualizing deep neural decisoin forest:

1. The classification models for MNIST and CIFAR-10 are released.
2. Pre-processed CACD dataset is released.
3. A model trained for CACD dataset is released, which achieved state-of-the-art accuracy and comsumes less memory.

Finally, this repository is version 1.0 with more comments to write.

Decision making for Image Classification

Decision making for Facial Age Estimation

Note how the irrelevant texture (e.g. hair) is ignored by the model during the decision making process.

Accuracy on Cross-Age Celebrity Dataset (CACD)

Model	Error	Memory Usage	FLOPs
DRFs (CVPR 2018)	4.637	539.4MB	16G
RNDF (Ours)	4.595	112.4MB	4G

Requirements

• Python 3.6 (not tested for other versions)
• PyTorch 1.0
• CUDA (Yet using CPU is possible if you want to modify the code)

Pre-trained models

You can download the pre-trained models at https://drive.google.com/drive/folders/1DM6wVSknkYBqGf1UwHQgJNUp40sYDMrv?usp=sharing and place them in the "pre-trained" folder.

Visualization for Image Classification

After downloading the pre-trained models, go to /src and simply run

python ndf_vis.py 

for CIFAR-10.

For MNIST, run

python ndf_vis.py -dataset 'mnist'

Visualization for Facial Age Estimation:

To visualize NDF for CACD dataset:

1. Download the pre-processed images at https://drive.google.com/file/d/1OBu62cpnaMl5EX8EsjfEenRVv9rk3trt/view?usp=sharing and decompress it into the "/src/data" folder.
2. Download metadata folder at https://drive.google.com/drive/folders/1s_Ml82O4FVkC34PCE4ttrYhta3EKeYdo?usp=sharing and place it under the "/src/data" folder.
3. Go to /src/age_estimation and run

python ndf_vis.py 

Sorry about the hard-coded paths and few comments. Please refer to the similar classification code for detailed comments for now. Future updates will introduce more comments.

Training for Image Classification

To train a deep neural decision forest for CIFAR-10, use

python main.py

For MNIST, run

python main.py -dataset 'mnist' -epochs 50

Training for Facial Age Estimation (Beta)

To train RNDF for CACD dataset: follow the same step 1 and 2 as in visualization for data preparation. Finally, go to /src/age_estimation and run

python main.py -train True

To test a pre-trained model for CACD, go to /src/age_estimation and run

python main.py -evaluate True -test_model_path "Your Path"

The released model should give a MAE of 4.59

License

MIT

Citation

Please consider citing the related papers in your publications if they help your research:

@InProceedings{Li_2019_CVPR_Workshops,
author = {Li, Shichao and Cheng, Kwang-Ting},
title = {Visualizing the Decision-making Process in Deep Neural Decision Forest},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}

@article{li2019facial,
  title={Facial age estimation by deep residual decision making},
  author={Li, Shichao and Cheng, Kwang-Ting},
  journal={arXiv preprint arXiv:1908.10737},
  year={2019}
}

@inproceedings{kontschieder2015deep,
  title={Deep neural decision forests},
  author={Kontschieder, Peter and Fiterau, Madalina and Criminisi, Antonio and Rota Bulo, Samuel},
  booktitle={Proceedings of the IEEE international conference on computer vision},
  pages={1467--1475},
  year={2015}
}

Links to the papers:

• Visualizing the decision-making process in deep neural decision forest
• Facial age estimation by deep residual decision making
• Deep neural decision forests