• Update tiling method for inference on large images

• Modify segmentation map weights
• Update empty tile detection method

• Output postprocessing settings (thresholds) in scoring results
• Allow marker image to be omitted from postprocessing
• Use all GPUs by default for inference if unspecified

• Updated postprocessing algorithm
• Addition of DeepLIIFExt model class

• Ignore empty tiles during inference.

• Stable model serialization and inference between runs
• Memory efficient inference function
• Option to pseudo-color marker and DAPI inferred images
• Use training options in test script

• VGG loss added to the translation task.
• Automatically reading the model parameters to create the model for inference.
• GPU ids configuration updated.
• Converting the images to 'RGB', fixing the issue of reading grayscale or 'RGBA' images.
• Saving model parameters in training mode added to the cli.py file.

Added features:

• Pip installation
• Client commands
• Multi-GPU training (DP and DDP)
• Model Serialization
• Deployment with Torchserve
• ImageJ Plugin