This project is a reproduction of the SIGGRAPH 2016 paper Automatic Triage for a Photo Series written in Python with the help of Keras and TensorFlow.
data/ # benchmark dataset (Princeton Adobe photo triage dataset)
| demos/ # three demo scenarios (jpeg files)
| download.sh # data downloading and preparation
src/ # source code
| data.py # data loading and preprocessing
| models.py # models with different settings
| train.py # script for training
| evaluate.py # script for evaluation
| predict.py # script for prediction
Python 2.7OpenCV 2Keras 2.0+TensorFlow 1.0+
cd data/ && sh ./download.shcd src/ && python train.py <options>--exp experiment identifier (default: default)
--gpu GPU used for training (default: 0)
--epochs number of training epochs (default: 16)
--batch mini-batch size (default: 4)
--model model (default: vgg16) (vgg16 | vgg19 | resnet50)
--siamese weight sharing (default: share) (share | separate)
--weights transfer learning (default: imagenet) (imagenet | random)
--module feature interaction (default: subtract) (subtract | bilinear | neural)
--activation activation function (default: tanh) (tanh | relu)
--regularizer regularizatiation function (default: l2) (l2 | none)
cd src/ && python evaluate.py <options>--exp experiment identifier (default: default)
--gpu GPU used for evaluation (default: 0)
cd src/ && python predict.py <options> <image-list>--exp experiment identifier (default: default)
--gpu GPU used for prediction (default: 0)
In order to produce the prediction for demo scenario 1, you may use the following command:
cd src/ && python predict.py ../data/demos/scenario-1/scenario-1-a.jpg ../data/demos/scenario-1/scenario-1-b.jpgAlso, you may use the following command for short:
cd src/ && python predict.py ../data/demos/scenario-1This project is released under the open-source MIT license.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent