[Frigate+] Replace existing camera (name) with different make/model?

[dseven]

Not sure if this is a "stupid question".....

I've been using Reolink 5MP cameras, and uploading images to Frigate+ to make a fine-tuned model. I've been a bit underwhelmed by the Reolink performance, especially for one scene in particular. I came across a deal on a new-in-box Hikvision (8MP, 1/1.8" sensor), and picked it up. The difference in quality is quite remarkable!

I want to replace my existing "yard" camera with the Hikvision (and would prefer not to use a different name for it). If I just modify my Frigate config accordingly, is this going to confuse the Frigate+ model somehow? Would it be better to "start over" with a different camera name? I don't know enough (anything, really) about how the models work internally...

[inkeep-ai-assistant[bot]]

Replacing your existing camera with a new make/model (while keeping the same camera name in Frigate) is possible, but there are some important points to consider regarding Frigate+ model training and image uploads:

• Frigate+ models are fine-tuned using the images you have uploaded and annotated for each camera(1). The model doesn't inherently know about the physical camera hardware, but the images you upload reflect the scene, angle, resolution, and image quality of the previous camera(1).
• If you replace the camera but keep the same camera name in Frigate, any new images you upload for that camera will be from the new device, which may have different characteristics (resolution, color, field of view, etc.) compared to the old camera. This can introduce diversity into your training set, but it can also reduce the effectiveness of fine-tuning if the new images differ significantly from the old ones(1).
• For best results, it is recommended to submit at least 100 verified images per camera, covering a variety of conditions (lighting, weather, etc.)(1). If you switch to a new camera, you should focus on uploading and annotating new images from the new camera so the model can adapt to its characteristics(1).
• You are not required to use a new camera name, but if the scene and image characteristics have changed significantly, starting with a fresh set of training images (and optionally a new camera name) can help keep your dataset consistent and improve model performance for the new hardware(1).

In summary: you can keep the same camera name, but be sure to upload and annotate new images from the replaced camera so your fine-tuned model learns the new scene and image characteristics. If you want to keep your training data clean and avoid mixing images from two very different cameras, consider using a new camera name for the new device(1).

📚 Sources:

1. Models

---

Have another question? Just tag @inkeep.

[NickM-27]

The AI is generally correct here, there is really no reason to rename the camera or do anything specific. I have changed cameras, sometimes even the mounting position, many times in this way without any issues.

[dseven]

Hmm. The AI response actually sortof makes me lean towards making a new name - "it can also reduce the effectiveness of fine-tuning if the new images differ significantly from the old ones", but I'm not sure what constitutes "differing significantly". The new camera has a wider field of view and higher resolution (for the substream used for detection) - but I suppose that maybe just means I need to train my model about objects in the peripheral areas not seen by the old camera. The images are significantly sharper on the Hikvision, which is what I was hoping for. I'm just wondering if the "blurry" images already uploaded from the Reolink would confuse anything. Perhaps not, since the Hikvision (hopefully) won't be producing such blurry images! OK, I think I've talked myself into keeping the existing name and will see how it goes. Thanks!

[NickM-27]

Making a new name won't change that behavior, it essentially just means you'll need to upload more images for this new camera. I have done this similarly like I said without any issues

[NickM-27]

Feel free to create a new discussion if something else comes up