ATSVisionTrainer Help Center¶
This manual reflects the current UI and parameter adapters, covering four modern architectures, two classic architectures and anomaly detection. Choose the task page first, then prepare a matching dataset; configurations from different architectures are not interchangeable.
| What you want to do | Read more |
|---|---|
| First training run | Quick start |
| Prepare detection, classification or oriented-box data | Dataset preparation |
| Locate objects with boxes | Modern detection, classic detection 2.7.0 |
| Classify a whole image | Modern classification, classic classification 2.5.0 |
| Output each object's pixel region | Instance segmentation |
| Output oriented object boxes | Oriented detection |
| Detect anomalies using only OK images | Anomaly detection |
| Review training results or free disk space | Training history |
| Check every parameter's default, range and meaning | Parameter reference |
Read this online manual directly in your browser without installing or starting the application. Help inside the application continues to provide the bundled offline version.
The manual is available in Simplified Chinese, English, Italian, Japanese and Traditional Chinese. Switch languages using the toolbar at the top of the page.
About defaults
Defaults in this manual are the current initial UI values, not assumed values for every model YAML. When a classic architecture uses its configuration, the final training values come from the selected model configuration. Use each experiment's effective configuration and raw log to verify actual behavior.