Object Classification Problems

[umtsti]

I have created new sublabel classification as "Employee classification" under "Classification" as below

I have selected like 100 images per person and trained model. For each employee, I have selected different angles from different cames as below

? Here is "Hüseyin Cam" labed objects as you can see below this with trained with 100 images which is shown on previous imge.

But result is always wrong for all employess. Am I doing something wrong

[inkeep-ai-assistant[bot]]

---

I cannot find specific troubleshooting steps for your exact scenario in the available sources. However, based on the object classification documentation, here are some important points that may help:

Object classification uses a two-step verification process before assigning a sub label(1):

1. Threshold requirement: Each classification attempt must have a confidence score that meets or exceeds the configured threshold (default: 0.8)(1)
2. Class consensus: After at least 3 classification attempts, 60% of attempts must agree on the same class label(1)

To debug your classification model, enable debug logging by adding this to your configuration(1):

logger:
  default: info
  logs:
    frigate.data_processing.real_time.custom_classification: debug

(1)

The debug logs will show classification probabilities for each attempt, whether scores meet the threshold requirement, consensus calculations and when assignments are made, and object classification history and weighted scores(1).

For improving the model, the documentation recommends(1):

• Keep classes visually distinct and relevant to the chosen object types(1)
• Use the model's Recent Classification tab to gather balanced examples across times of day, weather, and distances(1)
• Ensure examples reflect object crops similar to Frigate's boxes; keep the subject centered(1)
• Tune threshold per model to reduce false assignments, starting at 0.8 and adjusting based on validation(1)

You may also want to check if you have the optional save_attempts configuration set, which defines the number of classification attempts to save in the Recent Classifications tab (default is 200 for object classification models)(1).

**

📚 Sources:

1. Object Classification

---

Have another question? Just tag @inkeep.

[hawkeye217]

The main issue here is that object classification really isn't the right tool for identifying individual people, especially in the context of a workplace. Object classification trains on the full person bounding box - meaning the model is learning from clothing, background, body shape, and other contextual details that change constantly. Your results are consistently wrong because the model can't reliably distinguish between employees based on those features.

What you want is Frigate's built-in Face Recognition, which uses an ArcFace-based embedding model specifically designed for identifying individuals by their facial features.

[umtsti]

But my camera angles is ok for this tool ?

[hawkeye217]

It may not be ideal, but it would probably work much better than object classification.

Make sure you read the training tips and the official documentation. Correct training is critical to good results with face recognition.

[umtsti]

I think my cameras is too far from ground to detect objects. Beacuse since I enabled face detaction, total detected face is 3 for whole day for 20+ cameras