In detection mode, the model returns classes, frames and scores. The developer can run the ready weights or train the option for a marked sample.
First, find out what is considered an object.
For a warehouse camera exercise, asking to find boxes is only a starting point. Decide whether to count a partially obscured box or packaging in a reflection. Ultralytics YOLO detection outputs bounding boxes, classes, and scores. Match your task to the class list for the selected weights. If the desired object category is missing, you need your own labels and a separate evaluation after training.
Threshold affects two types of errors
Capture a few sample frames with small objects near the edge, look-alike objects, and an empty scene. Compare thresholds on the same set. A stricter threshold can remove false boxes while also hiding useful detections. Record both effects alongside the detection count. If the camera will operate under different lighting, hold out some frames from those conditions.
How to mark the control sample
Agree on frame rules for a partially visible object and fix how to mark several overlapping instances. The images from the same video are almost identical. Divide by record or date if you want to see behavior on new scenes. Before working on the implementation, check the computational delay, the selected login size and licensing. One good example shows the result format, but the conditions of use and reliability are checked separately. Fix the settings to repeat the measurements later. Keep several control frames next to the result.
Suitable tasks
- Check object recognition on test frames
- Compare gaps and extra frames at different thresholds
Strengths
- There are separate modes of prediction and learning
- Documentation describes inputs, thresholds and result formats
Limitations
- A pretrained model is limited to its own set of classes
- A confidence threshold cannot substitute for an evaluation set
Selection and conditions
- Free access
- Yes
- Trial period
- Not confirmed
- No sign-up required
- Not confirmed
- Russian interface
- Not confirmed
- Local installation
- Yes
- API
- Not confirmed
“Not confirmed” means there is not enough information. The feature may still be available; check the developer’s terms.
- Input formats
- Image, Video
- Result
- Data
Terms
Ultralytics publishes AGPL terms and offers a commercial license. Before adopting it, check the terms for the code and weights for your use case. A pretrained model is limited to its own set of classes.From task to first result
Practical example
- Task
- Check object recognition on test frames
- What to prepare
- Educational footage with known objects, similar objects and an empty stage.
Task brief or prompt
Run the detector on the same frames at different thresholds. Save classes, frame coordinates and scores.
Expected result
List of correct limits, omissions and extra triggers with indication of threshold.
How to check the result
- Look separately for rare classes and objects at the edge of the frame.
- Check that the training shots did not fall into the test sample.
An editorial teaching example of a task brief, not the result of hands-on testing of the service. Check that the required mode is available, that you have rights to the source material and that the intended use of the result is permitted.
Check details
Official sources
- Predictions with Ultralytics YOLO
- Training models Ultralytics YOLO
- Official Ultralytics Licensing Terms
This description is based on the developer’s material. Features, prices and availability may change; check the primary source before use.