Frigate+ Quick tips i found for labeling

[Fratopolis]

So I am sure most of us label images to the best of our abilities. I recently found out that labeling images on a 4k monitor and the resolution actually set to 4k with scaling set to 100% lets you move the bounding boxes even more finely especially zoomed all the way in. So I went through thousands and thousands of images tightening each and every bounding box. Opening each one independantely was painful but I found a way to open 100 of them at once then the next 100 and so on.....

Hope this info helps others

Anyone else have any tips on their labeling efforts?

[inkeep-ai-assistant[bot]]

---

Great tips on using higher resolution displays for more precise bounding box adjustments. Here are some additional best practices from the official documentation:

Core Labeling Guidelines:

• Label every object in the image - It's important to label all objects in each image before verifying. If you don't label a car for example, the model will be taught that part of the image is not a car and it will start to get confused(1)

• Make tight bounding boxes - Tighter bounding boxes improve recognition and ensure accurate bounding boxes are predicted at runtime(1)

• Label the full object even when occluded - If you have a person standing behind a car, label the full person even though a portion of their body may be hidden. This helps predict accurate bounding boxes and improves zone accuracy. If an object is partly out of frame, label it only when a person would reasonably be able to recognize the object from the visible parts(1)

• Label objects hard to identify as difficult - When objects are truly difficult to make out, such as a car barely visible through a bush, or a dog that is hard to distinguish from the background at night, flag it as 'difficult'(1)

Workflow Shortcuts:

Frigate+ provides keyboard shortcuts to speed up your workflow(1):

• w - Add box
• d - Toggle difficult
• s - Switch to the next label
• tab - Select next largest box
• del - Delete current box
• esc - Deselect/Cancel
• ← ↑ → ↓ - Move box
• Shift + ← ↑ → ↓ - Resize box
• scrollwheel - Zoom in/out
• f - Hide/show all but current box
• spacebar - Verify and save

AI Suggested Labels:

If you have an active Frigate+ subscription, new uploads will be scanned for the objects configured for your camera and you will see suggested labels as light blue boxes when annotating. These suggestions are processed via a queue and typically complete within a minute after uploading(1)

Suggestions are converted to labels when saving, so you should remove any errant suggestions and keep the most accurate bounding box, deleting any duplicates so you have just one label per object remaining(1)

Handling False Positives:

When you encounter misidentified objects, you should submit it as a false positive in Frigate and add a label for the correct object. The boxes will overlap(1)(12)

📚 Sources:

1. Annotating your images

---

Have another question? Just tag @inkeep.

[xconverge]

My tip is it sets your next box label type to the last box you clicked, so if you want to create a box for a person, it's faster to click a different person box first than it is to create a box and cycle through trying to find person, only to miss it and need to loop all the way around.

[NickM-27]

only to miss it and need to loop all the way around.

Just use shift + S to go backwards

[xBOHICAx]

I've been doing this. Has made things so much quicker

[xBOHICAx]

only to miss it and need to loop all the way around.

Just use shift + S to go backwards

Whoa

[xconverge]

only to miss it and need to loop all the way around.

Just use shift + S to go backwards

Whoa

😆 #20426