runprediction.sh

Task:

Main script to segment images using a trained model

Input arguments:

• a folder with a trained model (see CDeep3M modelzoo for pre-trained models or here how to train a model)
• a folder with images (sequential .png or .tif files with same x/y dimensions)
• the folder where to write the output

Example Command:

runpreditcion.sh trained_net_dir images_dir predict_outdir

Output directory:

ls $predict_outdir

1fm 3fm 5fm DONE augimages de_augmentation_info.json enhanced ensembled logs overlay package_processing_info.txt predict.config readme.txt

Content

enhanced: Autoenhanced images in CDeep3M2

ensembled: ensemble of predictions (normally of 1fm, 3fm and 5fm)

overlay: Predictions overlayed on top of the enhanced images

Additional Settings:

Models: --models 1fm,3fm,5fm

Augspeed: --augspeed 1 (options are: 1, 2, 4 or 10)

Denoising: --denoise -1

If you wish to turn of denoising e.g. in order to use models trained with CDeep3M 1.X, set this value to 0 in runprediction.sh: --denoise 0

Examples: runprediction.sh --denoise -1 trainoutdir imagesdir predictoutdir

runprediction.sh --denoise 0 trainoutdir imagesdir predictoutdir