View/Test Models

Test the performance of your trained model with Inference Jobs

Create Inference Job

In the model detail view, you can create "Inference Jobs" to test the model

• You have to deploy the model before creating any inference task.

• There are two modes of model inference: batch & real-time.

Real-time inference

Select and open one JPG/ PNG image from your local device, the inference result will be shown automatically with the AI image viewer.

The AI viewer contains the following information of each image:

Type	Image classification	Object detection	Object Segmentation
prediction results	label name & confidence score	label name & bounding boxes in dotted lines	label name & objects with white stripes
Explainability	heat map of each label (can be toggled on of by )	NA	NA
Ground truth (if available)	ground truth labels	Bounding boxes in bold lines	Objects painted in solid colors

Image Classification

Image Classification: prediction & ground truth

Image classification: heat map on

Object Detection

Object Segmentation

Batch inference

• Batch inference: Select one of the dataset that you have uploaded from the drop list, enter the inference name and complete the creation process.

  • Prediction only: For datasets without annotation data, inference result will only provide prediction results without an inference report.

View Inference Detail

Annotation Data (with ground truth/label)

• The inference detail, including training task, dataset, and annotation data will be shown here.

• Inference List: You can view the inference detail about the dataset, accuracy, and status.

• Inference Dataset shows all the cases that have been trained in this task.

Inference Detail for different types:

Type	Image Classification	Object Detection	Object Segmentation
Performance metrics	>Accuracy >ROC/PR Curve >Confusion Matrix >Evaluation Summary	>mAP >FROC/PR Curve >Evaluation Summary	>DICE scores >Evaluation Summary
Dataset Statistics	>Label Distribution	>Box Label >Box-per-Image >Box Ratio >Box Area >Label Distribution (Image-wise)	>Label Distribution >Mask-Coverage Distribution

Image Classification

Inference Detail: Image Classification

Object Detection

Object Segmentation

Prediction only

• if the batch inference dataset does not contain ground truth (annotation file not selected), there will be no performance metrics & Dataset statistics

• Each image of the selected dataset will be shown in the last column, you can check the result by clicking on the image.