This is the comment section where we take a look at what you have to say about Meta's SAM 3 model. It is an object recognition model able to track things in video, and they've just released SAM 3, but you've got to request access.
The setup is really simple - this is all of the code you need. You provide a URL to the video and then give it a text prompt like "track the dog" or "track the white football," and it can track that object through the video. The performance looks really, really, really performant, and you can also train the model as well.
To use it, you need to set it up in Python 3.12, clone the repo, and request access to the hugging face repo. The model is 0.9 billion parameters, and while I haven't gotten access yet, this looks very, very, very cool.
—
TIMESTAMPS:
0:00 - The Comment Section & SAM 3 🗣️
0:17 - What is SAM 3? (Object Tracking)
0:36 - The Simple Code Demo 💻
0:58 - Text Prompt Tracking Magic
1:22 - Python 3.12 Setup Instructions
2:16 - Hugging Face Access Request
3:10 - Model Size & Details 📊
3:23 - What is "CMD + AI"?
—
Unlock the full potential of your online presence with Kabarza and Samuel—experts in web design and development (respectively), powered by cutting-edge AI solutions. We blend creative design with advanced tech to deliver smart, high-impact websites that stand out. Ready to elevate your business? Contact us today and see what AI-driven innovation can do for you!
LINKS & RESOURCES:
Check out Kabarza's amazing work: https://kabarza.com
Visit Samuel's website for more: https://samuelgregory.co.uk
📷 Follow on Instagram: https://www.instagram.com/cmdaishow
—
HASHTAGS:
#Meta #SAM3 #AI #ObjectRecognition #ComputerVision #MetaAI #VideoTracking #MachineLearning #Python #HuggingFace
Transcript
This is the comment section where we take a look at what you have to say. Let us know what you think about anything you hear down in the comments and we might just read it out. We had a comment asking about what do we think of the SAM model. It is an object recognition model. It's able to to track things as a model and they've just released SAM 3. You got to request access which I'm still I was hoping to be able to go through this with you guys but I'm you got to request access. But let's go through the instructions and we can sort of talk talk a little bit about it. Really simple. This is the code. This is all of the code you need. And you provide it a URL to the video wherever it is. Uh video path here. And then you give it a text prompt and it can track that video whatever it is. Maybe it's track the dog. Uh track the white football white team or what player and white. There you go. and is even ignoring some of the some of the other players in white is because of that frame that first frame potentially um looks really really really performant and given that this is the this is the text and you can also train the model as well. So this is how to set it up. Cond is a a Python um version library. So you have you have to use it in Python 3.12. So you create an environment, a sandbox if you will of Python 3.12. Um, you I guess deactivate your current environment. You activate the SAM 3 environment which you just created. You install Torch and do all this that and the other. Clone the repo. You install it. And then I'm guessing somewhere here you need to basically request access to the hugging face repo. And if approved, you get five repositories which you can then link to. Oh, you can load an image. So I'm guessing that's maybe the image you want to say track this player or whatever. Don't know. I've got no need for this, but I would like to play with it. So, with all that being said, if I do get access to Hugging Face, I will absolutely uh try this out. I might even get access to it. It looks like it's an internal search tool. You have to um during the application process, you have to put in your affiliate. And I don't know, I'm not, you know, I just said on media. I'm just, you know, I don't know what else to to actually say. So, anyway, shame we couldn't get access to that. But this looks very very very cool. I wonder how big can we see how big the model is 0.9 billion. So what's that? That's 900,000 view license. Let's have a look here. Oh gosh. Unless require sound materials any output result there for a provider by as is. Uh, sounds it is. You'll have to have a look at this to see if it fits your, you know, your use case or whatever. Um, I'll try and update you on that as I can. Either way, SAM 3, it's a new one. I'll let you know. I'll update you when I know a little bit more about that. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.