Frigate Plus unverified images created for an older model vs newer model #20463
-
|
Hi all, Having a Frigate Plus account for a while, I have a lot of unverified images from the past, created when I was running an (very) older model (and detector) than I'm using now. Does it make sense to use the old images to train my much more newer model? Or is there a way to remove the older images from my Plus account, so I can start fresh again? Thanks for your advice. |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
Based on the available documentation, it's worth keeping and using your older unverified images for training your newer model. The documentation indicates that only images with at least one verified label will be used when training your model (1). This means your unverified images won't impact your model training until you verify them. When you verify and annotate these older images, they become valuable training data regardless of when they were originally captured (1). The key factors for effective training are image quality and diversity rather than when the images were created (2). The documentation emphasizes that submitting diverse images with varying conditions (cloudy days, sunny days, dawn, dusk, and night) improves results (2). Your older images likely contain this valuable diversity across different lighting and weather conditions. For the annotation process, you should label every object in each image before verifying (3). Make tight bounding boxes and label the full object even when occluded (3). The documentation recommends having at least 100 verified images per camera for very good results (2). Regarding removing older images, the available documentation doesn't provide information about deleting images from your Frigate+ account. For questions about account management features not covered in the documentation, you should visit GitHub or contact support directly. Would you like a general answer based on my knowledge instead, or do you need help with the annotation process for your existing images? 📚 Sources: Have another question? Just tag @inkeep. |
Beta Was this translation helpful? Give feedback.
-
|
The AI is correct. Any images you've submitted to Frigate+ and have verified are always used for model training, no matter what model type you select. Old images are indeed valuable training data - as seasons and lighting change, you're building a solid foundation for detection in all kinds of circumstances. |
Beta Was this translation helpful? Give feedback.
Based on the available documentation, it's worth keeping and using your older unverified images for training your newer model. The documentation indicates that only images with at least one verified label will be used when training your model (1). This means your unverified images won't impact your model training until you verify them.
When you verify and annotate these older images, they become valuable training data regardless of when they were originally captured (1). The key factors for effective training are image quality and diversity rather than when the images were created (2).
The documentation emphasizes that submitting diverse images with varying conditions (cloudy days, sunny …