5 Model Building
Below are the tools associated with model building in GRIME AI:
The Train Model tab allows for the building of automatic image
segmentation models using a ground truth annotation file (in COCO format) and
original images listed in the annotation file. Open the Deep Learning tab by clicking the 
symbol on the navigation bar.
The ML image Processing page will open.
- A user can enter a file path to a folder using the top navigation bar. The directory can have one or more (training) image folder. The following illustrates the recommended file structure.
- A (training) image folder must contain an annotation file and the original images listed in the annotation file in order to be detected by GRIME AI.
- GRIME AI searches recursively and all available image folders will be displayed on the Available Image Folders column. If you have selected the exact image folder that contains only an annotation file and images, a single period (".") may appear in the available image folders.
- The Image folders will appear in the Available Image Folders section. To use a folder of images, select the desired folder and click the Move to Right option.
- Before running the training, it is important to use the dropdown on the Label Selection to ensure it is set to the desired category.
- Then fill in the site Name
- Several training parameters can be adjusted on the leftmost column to influence the outcome of the training.
- When all desired images and parameters are set, select the Train button found underneath the Label Selection.
- A Training in-progress window will then open showing the percent progress.
- The new models can be found in a folder named ‘sam2’ under ‘Models’ within the GRIME-AI folder.
To begin using these new models see GRIME-AI Model Deployment