Plants found by their foliage, in 3–4 days
Upright
The plant side of a startup's warehouse inventory drones: a vision model that finds and counts potted plants, even from their foliage, built in 3–4 days.
LOG · WHAT IT IS AND WHAT IT DOES
Upright was building autonomous drones for inventory management in large warehouses. Its plant side is a vision pipeline: Grounding DINO finds what a text prompt describes in an image, SAM (Segment Anything) cuts out each one's exact outline, and a model trained on top recognizes the potted plants and counts them.
WHY
The founder wanted to try 2 routes: traditional warehouse crate inventory and barcode reading, vs. plants and pots. I had a little bit of vision background from working on Sophie 3, so I helped suggest some routes. Then I took on the plant side and the founder worked on the crate side, to compartmentalize, test both and pick a route.
WHAT I WORKED ON
I used Grounding DINO + SAM and trained a model to recognize plants, and I was able to get it to recognize them from their foliage too. It all took 3–4 days, while I was working on a gazillion other projects.
WHAT CAME OF IT
It was working, but the other route of warehouse drones was eventually chosen.
- OBJECT
- M40
- YEAR
- 2026
- CONTEXT
- Cornell Rev incubator
- CATEGORY
- work
- TYPE
- software
- STATE
- working
- TOPICS
PLATES · 3
PLATE IEight succulents, found and counted
PLATE IIEight in a half-empty tray, the empty pots skipped
PLATE III47 on a store's racks, at 25% confidence
MORE WORK SOFTWARE
See all 20
That was 1 of 50.