Roboflow

Build & use computer vision models fast ✨

Get started: source datasets & models

[Roboflow @roboflow]

No data, no problem

introducing agentic synthetic data generation with Cosmos 3

share a few examples, generate more data, automate model training, automatically deploy the latest version with no downtime

in a benchmark run with Corning Incorporated's optical fiber manufacturing engineering team, a model trained on 8 real defect images plus synthetic examples generated by Cosmos reached 0.95 mean average precision and perfect recall on the toughest defect class, beating a baseline trained on real data alone.

"The Roboflow Agent powered by NVIDIA allows us to generate the training data we need, fine-tune our models, and strengthen model performance and inspection quality while increasing the speed, scalability, and adoption of next-generation technologies,” - Jeremy Knopf, chief information officer, Corning Optical Communications

[NVIDIA AI @NVIDIAAI]

Introducing Cosmos 3: Our latest frontier model for Physical AI

Cosmos 3 is the world’s first fully open omnimodel with native vision reasoning, world and action generation.

Today we’re releasing Super (32B) and Nano (8B) variants.

[Roboflow @roboflow]

@mfranz_on @Tugrul_Guner @josephofiowa only the roboflow brand survives

[Roboflow @roboflow]

@Tugrul_Guner @mfranz_on @josephofiowa bingo

[Roboflow @roboflow]

@churchkey @mervenoyann @huggingface definitely! what type of coins are you thinking?

[Roboflow @roboflow]

@Tugrul_Guner @mfranz_on @josephofiowa other way around

[Roboflow @roboflow]

@_joe_harris_ well said, appreciate the share!

[Roboflow @roboflow]

@jorelikins glad we got it sorted!

[Roboflow @roboflow]

@aiseomastery From what we've seen, can be just 5 real examples. Totally depends on the environment, use case, defect types/variations, etc lots of variables

[Roboflow @roboflow]

No data, no problem

introducing agentic synthetic data generation with Cosmos 3

share a few examples, generate more data, automate model training, automatically deploy the latest version with no downtime

in a benchmark run with Corning Incorporated's optical fiber manufacturing engineering team, a model trained on 8 real defect images plus synthetic examples generated by Cosmos reached 0.95 mean average precision and perfect recall on the toughest defect class, beating a baseline trained on real data alone.

"The Roboflow Agent powered by NVIDIA allows us to generate the training data we need, fine-tune our models, and strengthen model performance and inspection quality while increasing the speed, scalability, and adoption of next-generation technologies,” - Jeremy Knopf, chief information officer, Corning Optical Communications

[NVIDIA AI @NVIDIAAI]

Introducing Cosmos 3: Our latest frontier model for Physical AI

Cosmos 3 is the world’s first fully open omnimodel with native vision reasoning, world and action generation.

Today we’re releasing Super (32B) and Nano (8B) variants.

[Roboflow @roboflow]

@aiseomastery depends on the type of defect but yes generally still does well

best way to operate is using active learning to continually update the dataset with new examples

[Roboflow @roboflow]

@mfranz_on crazy amount of potential here for sure

[Roboflow @roboflow]

good thread on using cosmos 3 for visual understanding and reasoning

https://t.co/X10wvj9SI9

[Erik Kokalj @erik_kokalj]

Tested NVIDIA's new Cosmos 3 myself. Easy tier first: zero-shot gate turnaround.
Single inference of 1.5min video, 0.5fps ingestion, 6 segmented states. Thinking on, no fine-tuning, prompted it to focus on cargo in the cropped ROI.
Harder tests below. 🧵

[Roboflow @roboflow]

and cosmos 3

https://t.co/0HUnSJoH5o

[Roboflow @roboflow]

@mfranz_on @BuzzyUzzy7729 @ASaltWeapons thanks for the shoutout. trying our best to make the universe experience better

would love to know any issues you have with it

[Roboflow @roboflow]

@clogwog working on it! ty

[Roboflow @roboflow]

@mervenoyann @lillyguisnet @huggingface if we all do our jobs right, agi will be embodied in a toaster

[Roboflow @roboflow]

rf-detr helping make big gains for small object tracking

detection rate from 47% to 86%

[Ruku @lohba_]

Swapped my ball tracker from TrackNet (3-frame temporal CNN) to RF-DETR (transformer, single-frame). Trained on ~4,700 labeled pickleball frames from public data. Detection rate jumped from 47% to 86%.

[Roboflow @roboflow]

@withcoreviz @mervenoyann @huggingface lets go!!