Use the GeoDict-AI module to train a neural network which will transform a low quality image into a high quality image. For example, scans may be taken from less angles or with reduced exposure time, if a network was trained to improve them. How to train a neural network to enhance gray value images is explained in the GeoDict-AI user guide.

If a suitable graphics card is detected during installation of GeoDict, the GPU mode is used for running Enhance Image (AI), otherwise it is running in CPU mode.

If multiple GPUs are available, select on which it should run. The image enhancement can also run on multiple GPUs. If selecting more GPUs than licensed, an error message appears when clicking Apply.

Browse for the neural network model trained with GeoDict-AI and best suitable for the image under consideration.

In the Description, the current constraints for the application of the neural network are listed, as entered in the GeoDict-AI module for training neural networks, e.g., the kind of images the neural network was trained for.

The choice for Batch Size is related to the memory available on the graphics card (GPU). Conceptually, GeoDict loads the graphics card with portions of work called batches. Currently, the selection must be made manually, and the parameter is set following the value entered in Batch Size. The higher the batch size, the better the performance. Thus, set the Batch Size as high as possible depending on your GPU and the settings of the neural network. If multiple GPUs are selected, the batches are distributed equally on the different GPUs.

If no GPU is detected, the image enhancement is run on the CPU, which needs much more runtime.

If the Batch Size is too large for what is available on the graphics card, an error message appears, and the image enhancement is aborted.

Click Apply to start the image enhancement.

Example

A carbonate sample was scanned once with short and once with long exposure time. The sample data was provided by Chandra and Al-Naimi. Using GeoDict-AI, a neural network was trained to transform the low quality scan into the high quality scan. Below find a comparison of applying the NLM Filter and applying the trained AI-model on the low quality scan, respectively. The NLM Filter already reduces noise, but the contrast is worse compared to the scan with long exposure time. The AI model, however, produces a scan with a contrast comparable to the high quality scan while having less noise. Once a model is trained with GeoDict-AI, short exposure times are sufficient to obtain high quality images using Enhance Image (AI).