Code release for the paper AEPI: Representation Learning and Evaluation of Human Ear Identification based on a blend of Residual Network and Spatial Encoding.
Authors:Usama Hasan ,Waqar Hussain,Nouman Rasool
In this work, we present automated ear identification model on Ear VN dataset, a large-scale ear images dataset in the wild.
- Multiple GPUs for training
All the codes are tested in the following environment:
- Linux (tested on Ubuntu 16.04/18.04)
- Python 3.6+
- PyTorch 1.6
a. Clone the AEPI repository.
git clone https://github.com/UsamaHasan/AEPI-Automated-Ear-Pinna-Identification
cd AEPI-Automated-Ear-Pinna-Identification && cd src
python train.py --epochs 100 --batch_size 256 --lr 1e-3
Method | Top 1 Accuracy | Top 3 Accuracy |
---|---|---|
VGG-19 | 55.34 | 72.13 |
VGG-19 + SE | 59.79 | 75.75 |
ResNet-50 | 60.55 | 76.64 |
ResNet-50 + SE | 66.22 | 81.54 |
ResNet-152+ SE | 75.5410 | 87.207 |
EarVN1.0: A new large-scale ear images dataset in the wild.
Hoang VT. EarVN1.0: A new large-scale ear images dataset in the wild. Data in Brief. 2019 Dec;27:104630. DOI: 10.1016/j.dib.2019.104630.