The easiest & fastest way to annotate images for vision AI. | Created by @wkentaro_ | 16K GitHub stars
Joined November 2023
- Tweets393
- Following2
- Followers340
- Likes1.3K
Labelme retweeted
annotation is turning into coding:
1. nail the spec on a few images
2. batch annotate
3. review, by you or another model
4. merge into your dataset
Labelme retweeted
same model, same effort. only the prompt changed.
the right prompt tells it to zoom in, leave out what covers the backpack, and use group_id for shapes with holes.
Labelme retweeted
Shape locking and Fit to Selection are now in Labelme. Both great for reviewing what you or your agents or AI models annotated.
pip install -U labelme
Labelme retweeted
left: opus5.5 medium, right: sol 6.1 low
thought opus is good at everything, but gpt seems unbeaten on this
Labelme retweeted
The new landing page is live
Messages are clearer, images are sharper, demos are engaging. It's just much better
It took a week even with swarm of agents but I love how it turned out. Hope you like it
labelme.io
Labelme retweeted
Three polygons, One plant
Group ID keeps disconnected parts together
Labelme retweeted
AI polygons can trace every bump in a boundary. Sometimes you want a smoother outline
Labelme v7.2:
- Adds control over AI polygon smoothness
- Groups disconnected parts as one object
- Puts shape colors in Settings
Plus fixes for Undo and cursor-centered zoom.
Labelme retweeted
Turning a mask into an editable polygon is a tradeoff
Too many points make it painful to edit. Too few erase useful detail
But you get to choose
Labelme retweeted
Ok, why not
Just listed labelme.io on outliked. Let me see how high it can go
Every like on this tweet push it to upper. Likes are appreciated
(Share your tweet so I can push yours too)
outlike.lol
Labelme retweeted
Change a config, it applies.
Quit the app, the config still there.
A save fails, your file and annotation intact.
v7.1 removes the surprises, so you can focus on the work.
Labelme retweeted
Cool app for Drone imagery annotation/labelling/reference data for CNN-based segmentation used ar
∆Paper1: for Multispectral imagery ~link.springer.com/article/10…
∆Paper2: LiDAR point clouds co-registered to RGB imagery ~eartharxiv.org/repository/vi… #computervision #deeplearning #nvidia – at Hobart, Tasmania
Labelme retweeted
Had to delete duplicates after every AI-Box. Now don't.
v6.3 collapses nested SAM masks. Skips objects you've already labeled.
Labelme retweeted
Just chatted with @lauransotomayor
I got to know her use case of @LabelMeAI on drone imagery to vegetation and more.
Cool to see the AI-assisted annotation working in there.
Labelme retweeted
Drag a box over a parking lot, get oriented rectangles. Drag one over a tank farm, get circles.
v6.2 ships both.
Labelme retweeted
Cars in a parking lot don’t line up with the X axis. Their bounding boxes shouldn’t either.
Oriented rectangles, landing in the next @LabelMeAI release.
Labelme retweeted
AI-Box gives you polygons. But sometimes the best fit is a circle.
@LabelMeAI is adding circle and rectangle to AI-assisted annotation. (Demoed on storage tanks.)