Marco Franzon's banner

Marco Franzon

@mfranz_on • 16,507 subscribers

LLMs · Computer vision · ESP32 · Agentic DataCenters @dualistic_twin

Shorts

This is the power of YOLO, trained on a laptop for ~1 hour, with a Kaggle dataset. Oh, and just ~100 lines of Python. I can make a startup on this and it took me literally a couple of hours.

This is the power of YOLO, trained on a laptop for ~1 hour, with a Kaggle dataset. Oh, and just ~100 lines of Python. I can make a startup on this and it took me literally a couple of hours.

4,873,361 次观看

Deepseek v4.1 Flash on 24GB Macbook. I think I reached my limit, 1.36 tok/s. It is far from being usable but the feeling of having a SOTA model on my consumer machine is awesome. I started from 0.74 tok/s and almost double it. It was a great journey.

Deepseek v4.1 Flash on 24GB Macbook. I think I reached my limit, 1.36 tok/s. It is far from being usable but the feeling of having a SOTA model on my consumer machine is awesome. I started from 0.74 tok/s and almost double it. It was a great journey.

90,881 次观看

I received a message in my direct messages asking how difficult it would be to use computer vision to track cell movement. > SAM zero shot for the immune cell > YOLO finetuned on a small dataset for bacteria Realization time: ~2 hours

I received a message in my direct messages asking how difficult it would be to use computer vision to track cell movement. > SAM zero shot for the immune cell > YOLO finetuned on a small dataset for bacteria Realization time: ~2 hours

188,673 次观看

Now you can make the so famous "pothole detector" using Ultralytics Platform as backend. All the compute is offloaded, you have just to make a cool UI to have your detector.

Now you can make the so famous "pothole detector" using Ultralytics Platform as backend. All the compute is offloaded, you have just to make a cool UI to have your detector.

159,497 次观看

The future of agriculture lies in optimization. A system like this costs very little: a drone with a camera and a computer vision system accurate enough to count and identify plants or diseases. It's so easy that anyone can make it at home for their own crops.

The future of agriculture lies in optimization. A system like this costs very little: a drone with a camera and a computer vision system accurate enough to count and identify plants or diseases. It's so easy that anyone can make it at home for their own crops.

120,683 次观看

I can't stop playing with Rapid Roboflow. The process is so smooth that you can just: - Go to the rapid roboflow platform - Drop a few images or a short video - Type a prompt like “yellow cap” or “person wearing red hat” - Instantly get a working detection model No training, no GPUs, no cloud setup. Plug it straight into your app. It’s also baked into Roboflow Workflows, so you can instantly build automations like “count vials crossing this line” in minutes. Computer vision just went from weeks in seconds.

I can't stop playing with Rapid Roboflow. The process is so smooth that you can just: - Go to the rapid roboflow platform - Drop a few images or a short video - Type a prompt like “yellow cap” or “person wearing red hat” - Instantly get a working detection model No training, no GPUs, no cloud setup. Plug it straight into your app. It’s also baked into Roboflow Workflows, so you can instantly build automations like “count vials crossing this line” in minutes. Computer vision just went from weeks in seconds.

176,108 次观看

Qwen3.6 26b is local AGI. Macbook pro m5 24gb ram. Backend: mlx-lm Tok/s: ~15

Qwen3.6 26b is local AGI. Macbook pro m5 24gb ram. Backend: mlx-lm Tok/s: ~15

69,564 次观看

X Hermes Agent desktop. I've added X into a side panel to have the full control over my agents while I am doomscrolling.

X Hermes Agent desktop. I've added X into a side panel to have the full control over my agents while I am doomscrolling.

44,892 次观看

Added an obstacle in the room and simulated three different scenarios. This makes any kind of airflow simulation inside a room much more accessible. You can optimize the ventilation of your living room with just 10 lines of Python.

Added an obstacle in the room and simulated three different scenarios. This makes any kind of airflow simulation inside a room much more accessible. You can optimize the ventilation of your living room with just 10 lines of Python.

142,636 次观看

Calculating pothole volume from a 2D image is challenging but possible: > Assign each pothole an ID to avoid duplication. > Use basic trigonometry to convert pixels to real-world meters. > Apply a novel algorithm to estimate depth using road engineering data.

Calculating pothole volume from a 2D image is challenging but possible: > Assign each pothole an ID to avoid duplication. > Use basic trigonometry to convert pixels to real-world meters. > Apply a novel algorithm to estimate depth using road engineering data.

143,993 次观看

The "hello world" was achieved. Soon everyone can build cool ESP32 projects starting from a prompt. And look the beautiful simulation.

The "hello world" was achieved. Soon everyone can build cool ESP32 projects starting from a prompt. And look the beautiful simulation.

66,436 次观看

If your datacenter hasn't a Digital Twin like this, what are you doing?

If your datacenter hasn't a Digital Twin like this, what are you doing?

25,810 次观看

Honestly the best case you can have for your DGX. - two exhaust fans underneath - tilted 30 degrees to improve ventilation - two side slides to keep it in place

Honestly the best case you can have for your DGX. - two exhaust fans underneath - tilted 30 degrees to improve ventilation - two side slides to keep it in place

25,085 次观看

wait... this looks 98% accurate

wait... this looks 98% accurate

47,856 次观看

This should be the airflow into a DGX Spark. That's why my cases are so good for it, because I study in detail their airflow.

This should be the airflow into a DGX Spark. That's why my cases are so good for it, because I study in detail their airflow.

19,336 次观看

Imagine choose a seat on the airplane after seeing this 3D simulation instead of that boring 2D free/occupied view. You can just make a startup using threejs and a good idea nowadays.

Imagine choose a seat on the airplane after seeing this 3D simulation instead of that boring 2D free/occupied view. You can just make a startup using threejs and a good idea nowadays.

19,528 次观看

Curling stone tracking v1 Olympic Games edition. How I made this first version: - YOLO for real-time stone detection - ByteTrack (Supervision) for consistent object tracking - SciPy for trajectory smoothing - OpenCV + NumPy for annotation, velocity, direction and acceleration visualization The colors indicate whether the stone is : accelerating → green stable → orange decelerating → red

Curling stone tracking v1 Olympic Games edition. How I made this first version: - YOLO for real-time stone detection - ByteTrack (Supervision) for consistent object tracking - SciPy for trajectory smoothing - OpenCV + NumPy for annotation, velocity, direction and acceleration visualization The colors indicate whether the stone is : accelerating → green stable → orange decelerating → red

21,382 次观看

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

I created a web app for teaching computer vision basics, allowing users to upload an image or video stream and instantly view it as a numerical pixel matrix, with options to display values in RGB, Hex, or Grayscale formats. It features predefined convolution kernels applied in real-time: - Sobel operator for edge detection (computes gradients to highlight boundaries) - Gaussian Blur: averages neighboring pixels with weights to reduce noise - Sharpen: boosts local contrast to enhance details This setup clearly visualizes how images are represented as matrices or tensors, how convolution slides kernels across pixels, and how operations like gradient computation, blurring, or sharpening modify the data step-by-step. I plan to use demos like this in my upcoming computer vision course to build intuition for fundamental image processing concepts, OpenCV filters, and CNN building blocks (e.g., why kernels extract features).

I created a web app for teaching computer vision basics, allowing users to upload an image or video stream and instantly view it as a numerical pixel matrix, with options to display values in RGB, Hex, or Grayscale formats. It features predefined convolution kernels applied in real-time: - Sobel operator for edge detection (computes gradients to highlight boundaries) - Gaussian Blur: averages neighboring pixels with weights to reduce noise - Sharpen: boosts local contrast to enhance details This setup clearly visualizes how images are represented as matrices or tensors, how convolution slides kernels across pixels, and how operations like gradient computation, blurring, or sharpening modify the data step-by-step. I plan to use demos like this in my upcoming computer vision course to build intuition for fundamental image processing concepts, OpenCV filters, and CNN building blocks (e.g., why kernels extract features).

271,121 次观看 • 9 个月前

Camouflage no longer works with YOLO. The detection of the hays here is impressive. YOLO can identify an hay which has mostly the same color of the background.

Camouflage no longer works with YOLO. The detection of the hays here is impressive. YOLO can identify an hay which has mostly the same color of the background.

149,665 次观看 • 6 个月前

Roboflow Rapid has revolutionized computer vision. You can train, deploy, and use a custom model with just one prompt and a video. I replicated my pothole detector in one minute on their platform by uploading an unlabeled video and writing "potholes".

Roboflow Rapid has revolutionized computer vision. You can train, deploy, and use a custom model with just one prompt and a video. I replicated my pothole detector in one minute on their platform by uploading an unlabeled video and writing "potholes".

195,470 次观看 • 10 个月前

I think I have made the coolest rig for an home AI setup. - 4× DGX Spark + OEM PSUs - 1× 8-port multi-gig switch - 8× short Cat6 - 1× vertical PDU / quality power strip - 3D printed chassis (and other small components + accessories) At the moment it is just a virtual replica. But if NVIDIA AI Developer wants to sponsor me with a real DGX Spark I am open to discuss

I think I have made the coolest rig for an home AI setup. - 4× DGX Spark + OEM PSUs - 1× 8-port multi-gig switch - 8× short Cat6 - 1× vertical PDU / quality power strip - 3D printed chassis (and other small components + accessories) At the moment it is just a virtual replica. But if NVIDIA AI Developer wants to sponsor me with a real DGX Spark I am open to discuss

44,590 次观看 • 2 个月前

This is MicroCAD. The programming language for your 3D projects. You can design your own Lego block in a few lines of code, ready to be exported as an STL file for 3D printing.

This is MicroCAD. The programming language for your 3D projects. You can design your own Lego block in a few lines of code, ready to be exported as an STL file for 3D printing.

113,663 次观看 • 10 个月前

The Future of gaming is awesome. No extra hardware required, you just need a normal pc and a camera to have a totally immersive experience. > mediapipe is going to be the next big thing

The Future of gaming is awesome. No extra hardware required, you just need a normal pc and a camera to have a totally immersive experience. > mediapipe is going to be the next big thing

104,709 次观看 • 10 个月前

I am not joking. This is a Navier-Stocks simulation with a DGX simplified geometry. What this solution means and how it is achieved? (the following part is for CFD lovers) D3Q19 lattice-Boltzmann, incompressible Navier-Stokes in the low-Mach limit, BGK collision with a Smagorinsky subgrid closure, full-way bounce-back on the voxelised chassis, plus a passive thermal scalar. 94×140×55 at 3 mm, 12000 steps, converged to residual 0.0013.

I am not joking. This is a Navier-Stocks simulation with a DGX simplified geometry. What this solution means and how it is achieved? (the following part is for CFD lovers) D3Q19 lattice-Boltzmann, incompressible Navier-Stokes in the low-Mach limit, BGK collision with a Smagorinsky subgrid closure, full-way bounce-back on the voxelised chassis, plus a passive thermal scalar. 94×140×55 at 3 mm, 12000 steps, converged to residual 0.0013.

32,073 次观看 • 2 个月前

I need to annotate some images for training a computer vision model. There are many powerful annotation platforms available, but I want to keep my images local. I added a new section to my CV Streamlit app to quickly annotate images and train a YOLO model in a few clicks.

I need to annotate some images for training a computer vision model. There are many powerful annotation platforms available, but I want to keep my images local. I added a new section to my CV Streamlit app to quickly annotate images and train a YOLO model in a few clicks.

29,530 次观看 • 9 个月前

You can build your 3D models directly from a prompt into Claude code. All thanks to this beautiful python project, build123 a parametric modeling framework for 2D and 3D CAD. It is built on the Open Cascade geometric kernel and it is suitable for 3D printing.

You can build your 3D models directly from a prompt into Claude code. All thanks to this beautiful python project, build123 a parametric modeling framework for 2D and 3D CAD. It is built on the Open Cascade geometric kernel and it is suitable for 3D printing.

20,915 次观看 • 6 个月前

I started to get passionate about the drones about 10 years ago: building, repairing (mostly), firmware tuning and so on. But now I am asking what if I had this app 10 years ago. This AI support for building from scratch a drone. >No days in modeling the frame on blender >No days in flashing bad firmware >No weeks for receiving the components, but 3D print them >No wrong components purchased A simple chat with the blender mcp for designing the drone and betaflight for test and simulate the firmware. I understand all of you who are building day and night, doing these things is too much fun.

I started to get passionate about the drones about 10 years ago: building, repairing (mostly), firmware tuning and so on. But now I am asking what if I had this app 10 years ago. This AI support for building from scratch a drone. >No days in modeling the frame on blender >No days in flashing bad firmware >No weeks for receiving the components, but 3D print them >No wrong components purchased A simple chat with the blender mcp for designing the drone and betaflight for test and simulate the firmware. I understand all of you who are building day and night, doing these things is too much fun.

15,417 次观看 • 4 个月前

I want to detect the gates to check how many were missed. But it is pretty complex: - Gates change size dramatically based on distance - No unique features: very generic shape, no distinctive texture or pattern - Model becomes biased toward detecting "gate-like" things

I want to detect the gates to check how many were missed. But it is pretty complex: - Gates change size dramatically based on distance - No unique features: very generic shape, no distinctive texture or pattern - Model becomes biased toward detecting "gate-like" things

12,957 次观看 • 8 个月前

没有更多内容可加载