Use cases

IBM® Maximo® Visual Inspection has uses cases that include static image classification, static image object detection, video object detection, static image segmentation, video action detection, auto label an image or video, and data augmentation.

Static image classification

Determine whether an image belongs to one or more classes of images based on overall image contents. For example, determine the species of bird in an image.

Figure 1. Detecting the overall contents of an image, based on custom training data
The image of a bird is next to a heat map, which is closer to red where the bird is located. It is blue further from the bird's center.

Static image object detection

Determine and label the contents of an image based on user-defined data labels. For example, find and label all black cars in an image.

Figure 2. Detecting and labeling instances of objects within an image based on custom training data
A picture of a black car is shown. A box is drawn around the box, and it is labeled as 'black car'.

Video object detection

Determine and label the contents of an uploaded video or live video stream based on user-defined data labels. For example, find and label all white cars in a video.

Figure 3. Detecting and labeling instances of objects within captured video frames based on custom training data
A picture of a white car is shown. There is a box drawn around the box and it is labeled as 'white car'.

Static image segmentation

Determine and label the precise location of objects in an image based on user-defined data labels and arbitrary shapes. For example, find and label the precise boundary of all leaves in an image.

Figure 4. Detecting and labeling the precise edges of an object within an image based on custom training data
An image of a leaf is shown with a line drawn around the edge of the leaf.

Anomaly detection

If you don't have images that show defects in an object, you can populate a data set with high-quality images of the object in a perfect state, without any defects. Train this data set with the anomaly optimized training type and the results highlight any anomalous regions in detected objects. You can then review the results to determine whether these anomalies are defects or not.

Table 1. Anomaly optimized model results
Element Description
Green bounding box The detected object
Red bounding box with dashed lines The anomalous region
Red point in the anomalous region The region in the anomaly that is considered to be the most anomalous
Result The confidence score for the anomalous region
Figure 5. A model trained to detect toothbrush heads showing anomolous regions with a red, dashed-line bounding box
A model trained to detect toothbrush heads showing anomalous regions with a red, dashed-line bounding box

Because model performance decreases as more objects are detected and depending on latency requirements, anomaly models are not suitable for situations in which multiple objects of interest appear in an image at the same time.

Video action detection

Annotate the parts of the video where a specific action is taking place. For example, detect a forehand or backhand stroke in a tennis game.

Auto label an image or video

After deploying a model for object detection, you can improve its accuracy by using the Auto label function. This function uses the labels in the deployed model to generate new labels in the data set, which increases the number of images that are labeled in the data set. The updated data set can be used to train a new, more accurate model.

By default, auto labeled tags are pink, and manually added tags are blue.

Figure 6. Auto labeled video
This image has several cars tagged with green boxes and blue boxes.

Data augmentation

After deploying a model, you can improve the model by using data augmentation to add modified images to the data set, then retrain the model. Data augmentation is the use of filters, such as blur and rotate, to create new versions of existing images or frames. Augmentation does not apply to full videos. It can be applied to a video’s captured frames in the same way that it is applied to images. When you use data augmentation, a new data set is created that contains all of the existing images plus the newly generated images, which are marked as augmented.

Figure 7. Augmented video
This image shows a video where each frame is duplicated. The original frame is intact. In the duplicate frames, the image is rotated.