shimeta_infer — Image Inference

Use shimeta_infer for single-image inference with custom or built-in models.

Basic Usage

shimeta_infer <model.zip> <image.jpg>

Using Custom Trained Models

Transfer the .xmm.zip from the training platform and the image to the board, then run with a single command:

shimeta_infer model.zip image.jpg

Example (Built-in YOLOv8n Demo Model)

# Detect objects in an image (COCO 80 classes: person/car/cat/dog, etc.)
shimeta_infer /mnt/yolov8n_demo.zip photo.jpg

# Too few detections? Lower the threshold
shimeta_infer /mnt/yolov8n_demo.zip photo.jpg 0.1

# Too many overlapping boxes? Increase NMS
shimeta_infer /mnt/yolov8n_demo.zip photo.jpg 0.25 0.7

Understanding Results

After running inference, you'll see output like:

============================================================
  Detection Results: 1 object
============================================================
#    x1     y1     x2     y2     Confidence  ClassID  ClassName
0    719    447    755    559    84.4%       0        person

  Inference time: 24.10 ms

  Result saved: photo.jpg_output.jpg

Column	Meaning
x1, y1	Top-left corner of the bounding box (original image coordinates, in pixels)
x2, y2	Bottom-right corner of the bounding box
Confidence	How certain the model is — >50% is usually reliable
ClassName	The detected object class

Next

• Real-time Camera Inference
• SVP Scene Detection