YOLO11: How to Train for Object Detection | Live Coding & Q&A (Sep 30)

Рет қаралды 7,074

Roboflow

Күн бұрын

Пікірлер: 21

@culture2355 Ай бұрын

Tutorial starts at 14:21

@OnuralpSEZER Ай бұрын

Thank you so much for stream 🎉😊

@fzgarcia Ай бұрын

Hello, I have some questions about neural network processing with YOLO... I have a scenario where I need to identify coffee boxes whose only difference is the color to identify one type of coffee as different from another. I have always believed that for performance reasons, network training converts images to grayscale to work on only 1 color channel and does the same when inferring new images. Is this understanding correct? Is it possible to train the models considering the colors, the 3 RGB channels? Or would the best option be to identify the objects (bounding boxes) and perform post-processing with OpenCV, for example, in the bouding box region to identify the closest color? Thanks

@wangyubin1106 Ай бұрын

you may turn off the color data augmentation. For coco object detection, the NN should neglect color info。

@Roboflow Ай бұрын

I think that most neural networks use all color channels. A cool example is my football AI project: kzbin.info/www/bejne/l3O5eH6lg7-Ei9k I think it's safe to assume that the model distinguishing players from referees and goalkeepers is largely based on the color of their uniforms. I think that a single-stage pipeline makes a lot of sense, especially as a POC version. A multi-stage solution also makes a lot of sense, but it is prone to errors due to, for example, changes in lighting.

@ramazanoguz-q4n Ай бұрын

I am working on a project and considering using Roboflow. In my project, I will classify the geometric shapes (triangle, quadrilateral, circle, zigzag, etc.) found on shoe soles. At this point, I will label them using object detection. What is your suggestion on this?

@Roboflow Ай бұрын

Sounds great! Any specific problems you are facing?

@ramazanoguz-q4n Ай бұрын

@@Roboflow I am considering labeling and classifying the geometric shapes found on shoe soles using the object detection method. I plan to use thousands of shoes for my project. At this point, would it be appropriate to use object detection for this purpose? Is nested labeling a suitable approach, such as labeling both a rectangle and a star inside it?

@B4l0ur Ай бұрын

How can I then deploy the trained model on roboflow pls ? It is said it is not supported and if i canhed to yolov8 there is a depedency issue : ultralytics==8.0.196 is required whereas ultralytics==8.3.2 is installed

@Roboflow Ай бұрын

We are still working on it ;) stay tuned

@felixkuria1250 Ай бұрын

Can I use Yolo11 for tracking

@14types Ай бұрын

I want to put a camera at the front door and identify visitors - my family/stranger. What is the best way to do this?

@Roboflow Ай бұрын

Make some photos. Create dataset (probably 100 images for start). Train a nano model. Deploy raspberry pi.

@14types Ай бұрын

@@Roboflow can yolo11 do this? I've just never seen Yolo distinguish people's faces. Usually it just distinguishes between a person/dog, etc. i work with esp32cam, not raspberry.

@14types Ай бұрын

@@Roboflow I will send the photos to the server and then pass them through the neural network

@Roboflow Ай бұрын

If you don't plan to run the inference locally, the code you'll run on device will be really simple.

@Roboflow Ай бұрын

It can as long as thee set of people you want to detect is small. YOLO is only efficient up until 80-100 classes, so it's not really applicable in large-scale face recognition systems, but it can work for small problems like this.