Admin Panel

The VisionLog Admin Panel (/admin) is a protected management console accessible only to users with the admin role. It provides a centralized interface for registering new identities into the face database and managing the recognition engine's training databases.

Access Control

The /admin route is role-protected. Attempting to access it without admin privileges redirects to an Unauthorized Access page.

Roles are defined per user in the backend authorized_users.json configuration file. Each user entry specifies a role, a realm (organization), and a list of resources (face databases) they have access to.

{
  "email": "example@domain.com",
  "role": "admin",
  "realm": "Organization_Name",
  "resources": ["org_employees"],
  "active_resources": ["org_employees"]
}

Management Console Overview

Once inside, the admin sees the Management Console header with a live status pill showing:

• Active Databases — currently loaded .pkl face database names
• Registered Face Count — total number of enrolled identities

A Refresh Cache button is also available in the header to instantly clear the recognizer's in-memory cache and reload it from disk — useful after manual file changes or new enrollments.

The console is divided into two tabs:

Tab 1: Face Registration

The Face Registration tab walks the admin through a 3-step wizard to register a new person into the system.

Step 1 — Select Folder

The admin selects a category folder (top-level organizational directory on the server) from a card grid. Each card shows:

• 📁 Folder name
• Number of people already in that folder

The admin can also create a new sub-folder inside the selected category directly from this step, without leaving the wizard.

Step 2 — Enter Identity Details

Note: Use underscores instead of spaces (e.g., John_Doe). Names are case-sensitive and must be consistent with other enrolled identities.

Step 3 — Upload Images

The admin uploads one or more face images for the new person via a drag-and-drop / file-browse area.

Once confirmed, the system calls the /api/faces/register endpoint, which:

1. Saves the uploaded images to the correct folder path: <category>/<sub_folder?>/<name>/
2. Extracts 512-dimensional ArcFace embeddings from each image
3. Averages embeddings across all samples for robustness
4. Updates the active face database (unless Save images only was checked)

Result messages:

• ✅ Success — Person registered with confirmation message
• ❌ Error — Displays reason (e.g., no face detected, duplicate name)

Tab 2: Database Training

The Database Training tab allows admins to build or rebuild face recognition databases (.pkl files) from the stored image folders. This is necessary when:

• New persons have been registered with Save images only
• Bulk image changes have been made directly on the server
• A completely fresh database needs to be created

Configuration Options

Mapping Strategies

Training Process

Clicking Start Rebuild triggers a POST to /api/admin/train with the selected configuration. The backend:

1. Scans the source folder for person images
2. Detects and extracts ArcFace embeddings for each face
3. Writes the averaged embeddings into the target .pkl database
4. Returns a result summary

Training result output:

Database Management

After training, all available .pkl databases are listed. Each entry shows:

• Database file name (<name>.pkl)
• Whether it is currently ACTIVE (highlighted)
• An ACTIVATE button to switch the live recognition engine to a different database

Activating a database calls /api/admin/databases/active and triggers a global state refresh across the application — the recognition engine immediately begins using the newly activated database for all subsequent operations.

Admin API Endpoints

The following backend API routes power the Admin Panel:

Workflow Example

Scenario: A new employee joins the organization and needs to be added to the attendance system.

1. Navigate to /admin (requires admin login)
2. Go to the Face Registration tab
3. Step 1 — Select the organization's category folder (e.g., Citrus_Informatics)
4. Step 2 — Enter the person's name (e.g., Alice_Sharma)
5. Step 3 — Upload 3–5 clear face photos → click Confirm Registration
6. Switch to the Database Training tab
7. Select the existing database (e.g., Citrus_employees) and the same source folder
8. Choose Individual mapping strategy
9. Click Start Rebuild — training completes and the database is updated
10. The new employee is immediately recognized in live attendance sessions

Troubleshooting

"Redirected to Unauthorized Access"

• Confirm your account has "role": "admin" in authorized_users.json
• Log out and log back in to refresh your session token

"No face detected" during registration

• Ensure the uploaded image has a clearly visible, front-facing face
• Minimum face size: 50 pixels
• Avoid dark, blurry, or heavily occluded images

Training returns 0 processed people

• Verify the source folder contains correctly organized sub-folders
• Ensure person sub-folders each contain at least one valid image
• Check that the selected Mapping Strategy matches your folder structure

Cache not updating after new enrollment

• Click Refresh Cache in the Management Console header
• This forces the recognition engine to reload the database from disk