Roboflow

New video: Using vision-language models (like Qwen) to autonomously navigate and interact with ui elements. See the demo by ML Engineer Matvei Popov from @roboflow here: youtu.be/Y8CR6IzLDaw

@manaspgandhi super helpful, thank you

double backspace for delete, b for bounding box, command click to move around/draw a bounding box I only discovered these after looking up the shortcuts page (partially just because I didn't read when I hovered the tool), so some hints for users every now and then would be awesome!

@roboflow I love your shortcuts

@manaspgandhi thanks! here you go, fun comparison ready for you to run playground.roboflow.com/object-detecti…

@roboflow Oh awesome, could you link it here so I can go check it out? Love the content you’ve been putting out btw

lots of debate about Claude Opus 3.7 and GPT-5.5 for coding if you're curious which is better for vision tasks, we ran them through our eval and you can test them head to head GPT-5.5 is a lot slower but accuracy for object detection is really strong Try it out playground.roboflow.com/object-detecti…

@manaspgandhi yeah gemini is in there as well we have 80+ custom evals but the playground is loaded with out of the box prompts to try. or you can upload your own images+prompts to test them

@roboflow and did you try out gemini at all

here's your sign that AI for vision is very early and you should be building things

@frankdenbowgoat got me, my bad

@roboflow 3.7?

As I presented snapthatpiece.com at the @AITinkerers event at the @Shopify office yesterday, I was surprised that at least two teams came to my demo table and shared their attempts at building the same tool, but with sam3 and gemini / other LLMs. Apparently, some tasks aren't solved with a prompt. I also spoke on switching from Ultralytics to @roboflow RF-DETR. Apparently, few are aware of the AGPL trap of the former and glorious Apache-2.0 of the latter.

@me_harshraj09 @opencvlibrary @AIatMeta @Apple awesome. share with us when you have a demo!

Next upgrade: Moving to RF-DETR segmentation for all players to get cleaner masks. After that, I'm building an LLM-powered coach assistant. 🤖 Huge thanks to @opencvlibrary , @roboflow, @AIatMeta and @Apple for providing such an incredible ecosystem for developers to build on.🙌🚀

Is it possible to achieve accurate 3D mesh and event tracking for basketball during chaotic real-game footage? I decided to push the limits and build a full computer vision pipeline from scratch in one week to find out. Here is the breakdown of how I built it (and the hurdles I hit): 🧵👇

@manaspgandhi ty

@roboflow Facts, the model is nuts

RF-DETR with the Apache 2.0 license is a really good deal

@peteskomoroch @arb8020 You’ll never see us doing this in our prompts. Raccoons are always relevant.

@arb8020 Why include raccoons in a list of magical creatures?

gpt-5.5 prompt for codex seems to have a duplicated line trying to get it to not talk about creatures? Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query. [...] Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query gh link: #L55" target="_blank" rel="nofollow noopener">github.com/openai/codex/b…

@CoachCryptos @frankdenbowgoat @ballersappai american football is tough, we just made a new guide, maybe something helpful here for you to pick it back up blog.roboflow.com/american-footb…

@frankdenbowgoat @ballersappai @roboflow this is dope! I've been trying to do something like this for football (american), but it was was a nightmare lol!

Hackathon @ballersappai project by Harsh: - team + jersey tracking (on the fly) - player tracking (even when they disappear) - event tracking - court homography - 3D player tracking (skeleton + mesh) Thanks to @roboflow for training infra! Sign up: realballers.com

the level of fidelity and craft going into understanding sports at a deeper level is getting crazy

@marcel_butucea @thispatdeschere give it a try and let us know? trackers.roboflow.com/latest/tracker…

@thispatdeschere @roboflow SORT's Kalman + IOU works fine until occlusions dominate; OC‑SORT's trajectory‑aware re‑association slashes identity switches in crowded scenes. Curious how much extra latency it adds on edge devices? 🤔

This week's live session: How to track objects in video with SORT and OC-SORT algorithms, presented by Lee Clement, Machine Learning Engineer at @roboflow. Come hang out with us: luma.com/pd4hodix

for anyone interested in giving it a try mcp.roboflow.com

Roboflow MCP is starting to take shape thanks to the early folks using and giving feedback. Still limited scope but good functionality available now here's a quick demo showing you can build vision projects and use roboflow with your....voice more to come, appreciate any recommendations for what we should add next

great step by step view into creating a vision application by one of the most creative builders in AI

Come meet the @roboflow RF-DETR team at @iclr_conf this afternoon: Pavilion 4, Poster 3616!