IACV Project

Anomaly Detection Framework using Autoencoder and Denoising Network

Authors: Şemsi Yiğit Özgümüş, Yiğit Yusuf Pilavcı

Instructions

• If you don’t have the data folder, in the first run model will download and create the dataset.
• All the experiment configurations and model parameters can be changed from the related config files.

• To create the same environment used in the project:

conda create --name myenv --file spec-file.txt

• To run the model:

python3 train.py -c ./configs/\<CONFIGFILE\> -e \<EXPERIMENTNAME\>

• You can also use the same experiment name and configuration file to continue unfinished experiment.
• Since it’s a tensorflow based project, changes that affect the computation graph will result in failure to load the model. However you can modify the test_epoch() function to gain more insight about the model’s predictions. To make additional predictions without training the model and by loading it :

  python3 evaluate.py -c ./configs/\<CONFIGFILE\> -e \<EXPERIMENTNAME\>