Project Goal: The project aims to study on below topics/problems:
1. If a new image is added to a folder, is it causing anomaly in the existing dataset? e.g. training dataset is already labeled
2. IF a model is trained with a set of images can it detect contamination/anomaly in another folder?
3. Cluster images by their feature set.
Run as::
python driver.py
[arguments] [default] [options]
--backbone resnet resnet/vgg
--dataset food5k food5k/PascalVOC
--task anomaly anomaly/cluster
--subtask novelty novelty/outlier
Pre-Requisites:
- Python 3.6
- Tensorflow
- Keras
- sklearn
Note: It is assumed that datasets used in this projects are downloaded into "images" folder. Will update the code to get the dataset online and prepare the folder structures.
Methodology: