Should one add the full list of Object Tracks in the config file for each camera?

[sharpcall]

What is the need for this list? https://docs.frigate.video/configuration/objects/
If this list is not put there then the system will not detect them if they move? or will detect them if they move?
Is this list so that labels are put onto the object and otherwise no labels will show on the objects?

I am confused. I thought the system is to by default lable anything it see to the best of it's abilities. If I am putting a label of bike there only then the system will record the bike?

I read on that link page that adding these will help with lowering the overhead. How does that work? The goal is to capture anything useful, funny, unexpected, dangerous, safety, security, etc...AND keep the storage usage low. So how to achieve this balance?

[inkeep-ai-assistant[bot]]

You do not need to add the full list of object tracks for each camera unless you want Frigate to detect and track those specific object types. By default, Frigate only tracks certain objects (such as person), and will only detect additional objects if you explicitly add them to the objects.track list in your config(1). For example, if you want to track both persons and cars, you would add:

objects:  
  track:  
    - person  
    - car  

(1)

If you do not include an object in the track list, Frigate will not detect or track it—even if it appears in the frame(1). This list is not just for labeling; it controls what objects are actively detected and tracked. To balance between capturing interesting events and minimizing storage/processing overhead, only include the object types you care about. Tracking fewer objects reduces compute requirements and false positives(1).

If your goal is to capture a wide variety of events but keep storage usage low, you can configure Frigate to track only the most important object types and use features like zones, alerts, and retention policies to further filter and manage what gets recorded and stored(2)(1).

📚 Sources:

1. Review
2. Full Reference Config

---

Have another question? Just tag @inkeep.

[sharpcall]

Thanks.

Is that link a full list or there are other lists as well that can be added?

And where is the training modules set and what do they do? can they for example be fed previous history of dog pooping and be expected to recognize that for next time and not record the dog but record dog pooping?

[hawkeye217]

Yes, that link is the full list. The free model that ships with Frigate is trained on the COCO dataset. It can be retrained, but not without dedicated hardware and knowledge of object detection models.

Custom object and state classification ("is this my dog or someone else's dog" and "is the gate open or closed?") is a feature coming in Frigate 0.17.