AI Training#

1. Document Information (must be first)#

• Manual Version: v1.0
• Updated Date: 2026-03-02
• Applicable Software Version: AIBox Trainer 1.x
• Scope: Web console (Chrome/Edge/Firefox)
• Audience: End users / Administrators / Ops
• Conventions:
  • Buttons: 【Button】 e.g., 【Create】
  • Menus: “Menu/Submenu” e.g., “Datasets/Create”
  • Icons: described as (icon) e.g., (search), (delete)

---

2. Product Overview#

2.1 What it is and what problem it solves#

AIBox Trainer is an end‑to‑end object detection training platform covering media management, annotation, datasets, training, inference, and model management.

2.2 Main modules#

• Media: upload images/videos, register RTSP, keyframe extraction
• Annotation: online labeling with review
• Datasets: split, import/export, statistics
• Training: create training jobs with real‑time logs and metrics
• Inference: single image/video/RTSP inference
• Model Hub: manage, compare, export models
• Users & Permissions: role‑based access and quotas

2.3 Use cases#

• Industrial inspection: train defect detectors from historical images
• Security: live RTSP inference
• Outsourcing: centralized labeling and dataset generation

---

3. Runtime Environment & Preparation#

• Supported browsers: Chrome / Edge / Firefox (latest)
• Network: stable LAN/WAN (recommended > 20 Mbps)
• Resolution: 1920×1080 or above
• Accounts: created by administrators
• Roles:
  • admin: full access and system management
  • annotator: upload, annotation, training, inference
  • viewer: read‑only

---

4. Login & Logout#

4.1 Access URL#

Provided by ops, e.g., http://geoai-trainer.your-domain.com/

4.2 Login steps#

1. Open the URL

2. Enter username and password

3. Click 【Login】

4.3 Forgot password / reset#

• Contact admin to reset

4.4 Logout / timeout#

• Use the avatar menu 【Logout】
• Long idle time may trigger automatic logout

---

5. UI Overview (one‑page guide)#

5.1 Layout#

• Top bar: product name, language/theme switch, user menu

• Left: navigation

• Main: content area

5.2 Navigation & common buttons#

• Navigation: Dashboard, Media, Annotation, Datasets, Training, Inference, Models, Jobs, Admin
• Common buttons: 【Create】【Upload】【Export】【Delete】【Refresh】

5.3 Common operations#

• Create: click 【Create】 and submit the form
• Edit: use row actions (edit)
• Delete: click (delete) and confirm
• Search: use search box or filters
• Export: click 【Export】

---

6. Core Operations (main section)#

Each module follows the same structure

6.1 Dashboard#

• Purpose: system stats and task overview
• Entry: Navigation 【Dashboard】
• Steps:
  1. Open Dashboard
  2. Review cards and charts
• Notes / FAQs:

  • Statistics may be slightly delayed

6.2 Media Management#

• Purpose: manage images/videos/RTSP
• Entry: Navigation 【Media】
• Steps:
  1. Click 【Upload】 and select files
  2. Check items in the list
  3. Extract keyframes if needed
• Notes / FAQs:

  • Large files require stable network

6.3 Annotation#

• Purpose: draw boxes and assign classes
• Entry: Navigation 【Annotation】
• Steps:
  1. Open an image in the editor
  2. Draw boxes and select class
  3. Click 【Save】
• Notes / FAQs:

  • Class list is predefined by dataset/admin

6.4 Datasets#

• Purpose: build datasets for training and manage labels
• Entry: Navigation 【Datasets】
• Steps:
  1. Click 【Create】
  2. Link media or import data
  3. Split into train/val/test
  4. Click 【Export】 when needed
• Notes / FAQs:

  • Export format is YOLO

6.5 Training#

• Purpose: create and run YOLO training jobs
• Entry: Navigation 【Training】
• Steps:
  1. Click 【Create Training】
  2. Choose dataset, model, hyperparameters
  3. Click 【Start】
  4. Review logs and metrics
• Notes / FAQs:

  • Insufficient GPU may delay or fail jobs

6.6 Training Detail#

• Purpose: inspect logs, metrics, charts
• Entry: click a training job from the list
• Steps:
  1. Open job details
  2. Review real‑time logs and charts
• Notes / FAQs:

  • Requires stable WebSocket connection

6.7 Inference#

• Purpose: run inference on image/video/RTSP
• Entry: Navigation 【Inference】
• Steps:
  1. Select model
  2. Upload media or set RTSP
  3. Click 【Start Inference】
• Notes / FAQs:

  • Ensure RTSP URL is reachable

6.8 Model Management#

• Purpose: manage trained models
• Entry: Navigation 【Models】
• Steps:
  1. View model list
  2. Click 【Export】 to download
  3. Run inference test if needed
• Notes / FAQs:

  • Large exports may take time

6.9 Jobs#

• Purpose: view async job status
• Entry: Navigation 【Jobs】
• Steps:
  1. Open job list
  2. Filter by status
• Notes / FAQs:

  • Failed jobs show error details

---

7. Admin / Backend Functions#

7.1 User management#

• Create users, disable users, assign roles

7.2 Permission configuration#

• Role‑based access to modules

7.3 System settings#

• Storage quota, concurrency, system parameters

7.4 Logs / Data management#

• View system logs and job history

---

8. FAQ#

• Cannot login? Check credentials or contact admin to reset
• Page won’t load / slow? Check network and clear cache
• No response / no data? Refresh or retry later
• Export / upload failed? Check file size and network stability
• Insufficient permissions? Contact admin

---

9. Appendix#

9.1 Glossary#

• RTSP: Real‑Time Streaming Protocol
• YOLO: Object detection model family

9.2 Shortcuts / common actions#

• Ctrl/⌘ + F: search
• Ctrl/⌘ + R: refresh

9.3 Contact#

• Support: support@geotwin.cn
• Tech: tech@geotwin.cn