HCMV Capsid Detection using Tensorflow

Detection of HCMV capsid envelopment stages using Faster R-CNN Feature Extractor: ResNet101

Requirements

Python 3.5 or above Tensorflow 1.14.0 and above Install all the required dependencies stated in the requirements.txt file Best to run on a GPU

Dataset

Please use your own Electron Microscopy image dataset. Images used in this project will be provided upon request. For image request, please contact: clarissa.read@uni-ulm.de or jens.voneinem@uniklinik-ulm.de For questions regarding the codes please contact: kavitha.shaga-devan@uni-ulm.de

Usage

1. Fork and clone this repository.

2. Generate synthetic images

• Change the current working directory into the SinGAN folder
• Install all the required dependencies stated in the requirements.txt file in the SinGAN folder
• To train SinGAN model on your own image, put the desire training image under Input/Images, and run

'python main_train.py --input_name <input_file_name>'

• Please refer to https://github.com/tamarott/SinGAN for more information on the generation of synthetic images

3. Add your custom images

• Ensure that your current working directory is the main repository
• Your custom images should ideally have jpg extensions
• Annotate your images using any annotation tool that generate xml files
• Train/test split the image files into two directories, ./data/images/train and ./data/images/test according to the desired train/test split ratio
• Store the corresponding xml files inside ./data/images/train and ./data/images/test folders
• Commit and push your annotated images and xml files (./data/images/train and ./data/images/test) to your forked repository

4. Run Colab file

• Open 'tensorflow_object_detection_training_colab.ipynb' notebook in colab
• Replace the repository's url to yours and run it
• You can also opt to run this file locally in Jupyter notebook

The synthetic image generator in this repository has been obtained from https://github.com/tamarott/SinGAN.