Models and supported functions

You can train several types of models in IBM® Maximo® Visual Inspection.

Table 1 shows the training models that Maximo Visual Inspection supports and the types of training each one does.

Table 1. Model types
Model type Description Image classification Object detection Action detection
GoogLeNet System default when training for image classification. Yes No No
Faster R-CNN Optimized for accuracy. System default when training for object detection. No Yes No
YOLO v3 Optimized for speed. These models use you only look once (YOLO) v3 and take longer to train than other models in this product. Generates a Core ML file. No Yes No
Tiny YOLO v3 Uses the same technique as YOLO v3. Models are faster, less accurate, and use less GPU memory. Alternatively, these models can be deployed on the CPU. Generates a Core ML file. No Yes No
Detectron2 Detectron2 Mask R-CNN models can use objects that are labeled with polygons for greater training accuracy. You can disable segmentation for a shorter training time. No Yes No
High resolution Optimized for accuracy. Suitable for training and inference on high-resolution images. Supports segmentation to allow small objects, such as fine cracks, to be detected. No Yes No
Single shot detector (SSD) Used for real-time inference and embedded devices. It is almost as fast as YOLO but not as accurate as Faster R-CNN. No Yes No
Structured segment network (SSN) Used for video action detection. No No Yes
Anomaly optimized Optimized to identify anomalous objects. No Yes No
Custom model

Imported model used for training.

 Yes Yes No
Note: Starting in Maximo Visual Inspection 8.7, custom models are not supported. Custom models are still supported in Maximo Visual Inspection 8.6 and earlier versions.

Table 2 shows the functions that are supported for each model type.

Table 2. Supported functions
Model type Deploy multiple models to one GPU Deploy to TensorRT Enable Core ML Import and export model Supported on IBM Maximo Visual Inspection Edge Use for transfer learning TensorFlow Lite
GoogLeNet Yes

Yes

Version 8.8+ only

 Yes Yes Yes Yes No
Faster R-CNN Yes Yes No Yes Yes Yes No
YOLO v3 Yes

Yes

Version 8.8+ only

 Yes Yes Yes Yes Yes
Tiny YOLO v3 Yes

Yes

Version 8.8+ only

 Yes Yes Yes

Yes

Version 8.9+ only

 Yes
Detectron2 Yes No No Yes Yes Yes No
High resolution Yes No No Yes Yes Yes No
Single shot detector (SSD) Yes Yes No Yes Yes Yes No
Structured segment network (SSN) No No No Yes No No No
Anomaly optimized Yes No No Yes Yes No No

Custom model

 No No No Yes No No No
Note: Starting in Maximo Visual Inspection 8.7, custom models are not supported. Custom models are still supported in Maximo Visual Inspection 8.6 and earlier versions.