Welcome to MicroROS V2 Robot Car repository

2. CLI Usage

CLI Usage

1. Experiment Objective

Learn the basic usage of the YOLO command-line tool and independently test the image and video inference capabilities of YOLO models outside of ROS2.

2. Experiment Principle

1. Glossary

Term	Description
YOLO CLI	The command-line inference tool provided by Ultralytics, invoked with the `yolo <task> <mode>` format; no Python code is needed
yolo11	The Ultralytics YOLO11 series of models, lightweight and efficient, suitable for edge-device deployment
yolo26	The Ultralytics YOLO26 series of models, a next-generation architecture with higher accuracy
Confidence (conf)	The minimum confidence threshold for detection boxes; detections below this value are discarded
IoU (iou)	Intersection over Union, the overlap-area threshold used for deduplication in NMS

2. Technical Notes

The YOLO CLI is the command-line interface built into the ultralytics library. It supports all YOLO task types (detect, segment, classify, pose, obb) and modes (train, val, predict, export).

This section tests model inference independently outside of ROS2; it does not depend on micro_ros_agent, ROS2 nodes, or ESP32 hardware. You only need to activate the YOLO virtual environment to run it.

3. Experiment Steps

1. Hardware Requirements

Hardware	Requirement
PTZ version	✅ Supported
No-PTZ version	✅ Supported
Required peripheral	None (can be tested offline)

2. Activate the YOLO Environment


xxxxxxxxxx
yolo_env

3. Image Prediction

yolo11


xxxxxxxxxx
yolo detect predict \
  model=~/MODELS/yolo/yolo11n.onnx \
  source=~/MODELS/yolo/assets/bus.jpg \
  project=~/MODELS/yolo/assets/output \
  exist_ok=True \
  save=True \
  imgsz=640 \
  conf=0.25 \
  iou=0.7 \
  max_det=300 \
  device=cpu \
  show=False

Generated image path: /home/yahboom/MoDELS/yolo/assets/output/predict The terminal will print the generated path

yolo26


xxxxxxxxxx
yolo detect predict \
  model=~/MODELS/yolo/yolo26n.pt \
  source=~/MODELS/yolo/assets/bus.jpg \
  project=~/MODELS/yolo/assets/output \
  exist_ok=True \
  save=True \
  imgsz=640 \
  conf=0.25 \
  iou=0.7 \
  max_det=300 \
  device=cpu \
  show=False

The terminal will print the generated path

4. Video Prediction

yolo11


xxxxxxxxxx
yolo detect predict \
  model=~/MODELS/yolo/yolo11n.pt \
  source=~/MODELS/yolo/videos/car.mp4 \
  project=~/MODELS/yolo/videos/output \
  name=visdrone_predict_yolo11n_vehicle_conf003 \
  exist_ok=True \
  save=True \
  imgsz=960 \
  conf=0.03 \
  iou=0.75 \
  max_det=1500 \
  device=cpu \
  vid_stride=1 \
  classes=2,3,5,7 \
  show=False

The terminal will print the generated path

image-yolo11-detect-video-cli

image-yolo11-detect-video-result

yolo26


xxxxxxxxxx
yolo detect predict \
  model=~/MODELS/yolo/yolo26n.pt \
  source=~/MODELS/yolo/videos/car.mp4 \
  project=~/MODELS/yolo/videos/output \
  name=visdrone_predict_yolo26n_vehicle_conf003 \
  exist_ok=True \
  save=True \
  imgsz=960 \
  conf=0.03 \
  iou=0.75 \
  max_det=1500 \
  device=cpu \
  vid_stride=1 \
  classes=2,3,5,7 \
  show=False

The terminal will print the generated path

image-yolo26-detect-video-cli

image-yolo26-detect-video-result

5. CLI Parameter Quick Reference

Parameter	Description	Common Values
`model`	Model file path	`yolo11n.pt` / `yolo26n.pt`
`source`	Input source (image/video path)	File path
`project`	Output project directory	Directory path
`name`	Output subdirectory name	Custom name
`exist_ok`	Allow overwriting existing outputs	`True`
`save`	Save output results	`True`
`imgsz`	Inference input image size	`640` / `960`
`conf`	Confidence threshold	`0.03` ~ `0.25`
`iou`	NMS IoU threshold	`0.45` ~ `0.75`
`max_det`	Maximum number of detected objects	`300` / `1500`
`device`	Inference device	`cpu` / `cuda:0`
`show`	Whether to show the inference window	`False` (CLI mode)
`vid_stride`	Video frame-skipping interval	`1`
`classes`	Filter specific class IDs	`2,3,5,7`

4. Experimental Phenomena

Image prediction: The terminal prints the number of detected objects, and the result images are saved to the project/name/ directory.
Video prediction: The terminal prints processing progress and FPS frame by frame, and the result video is saved to the project/name/ directory.
Compared with yolo11, yolo26 usually has higher accuracy in the same scene, but inference takes slightly longer.

5. Common Problems

Problem	Cause	Solution
Model file not found	The model path does not exist	Check whether there is a `.pt` file under `~/MODELS/yolo/`
Output image/video is empty	Confidence threshold too high or `save=False`	Check `save=True` and lower `conf`
Inference is very slow	CPU inference or the model is too large	Use the lightweight `yolo11n` model and lower `imgsz`

YOLO Environment Setup

Object Detection