6 Model Deployment

Below are the tools associated with model deployment in GRIME AI:

• Segment Images
• COCO Generation

Segment Images

The Segment Images tab allows for automatic segmentation of images using locally stored machine learning models. Currently two model related file formats are supported: .torch and .pth

1) Choose a Model

• Under Select Model, click Browse and select a trained model file (e.g., a .torch file).
• The selected file path will appear in the Select Model field.

2) Select a Folder of Images

• Under Select Folder with Images, click Browse and select the folder containing images you want to segment.
• Select the folder, not individual images.

3) Select the Target Label

• In the Labels panel (right side), choose the label/class you want to segment (e.g., water, staff).
• The selected label determines which class the model will generate masks for.

4) Configure Output Options (Optional)

Under Output Options, choose what to save during inference:

• Save Predicted Masks - saves the generated masks for each image.
• Save Probability Maps - saves probability/confidence outputs (if available for the model).
• Copy Original Images - copies the original images alongside outputs for easier packaging/review.

5) Run Segmentation

• Click Segment Images to start processing the selected folder.

Model Metadata

The Model Metadata panel (bottom-right) displays helpful information about the selected model, such as:

• available categories/labels,
• training settings (e.g., learning rate, epochs),
• validation metrics (e.g., val loss, accuracy, mIoU),
• target category name and base model.

COCO Generation

The COCO 1.0 Generator tab allows users to generate COCO-format annotation files from existing images and mask files. These COCO files can then be used for model training within GRIME-AI or external deep learning frameworks.

1) Select Image Folder

• Click Browse next to Enter folder path for images (and masks)....
• Select the folder containing the images to be annotated.
• If masks are stored within the same directory, they may be detected automatically.

2) Select Mask File or Folder

• Click Browse next to Enter mask file path....
• Select the corresponding mask file or mask directory.

3) Use Single Mask File (Optional)

• Enable Use Single Mask File if all annotations are stored within one combined mask file.
• Leave unchecked if each image has its own corresponding mask file.

4) Generate COCO Annotations

• Click Generate COCO Annotations to create the COCO .json annotation file.
• The generated file can be used in the Train Model tab or exported for use in other machine learning workflows.

Notes

• Ensure image filenames match their corresponding mask filenames.
• Binary masks should follow consistent labeling conventions before generating COCO files.
• COCO output follows the standard COCO dataset structure (images, annotations, categories).

Back to Top