Google Colab haven't used for this, but you can read how to train YOLOv8 using dataset from Roboflow in the article for which this video had been recorded: dev.to/andreygermanov/how-to-implement-instance-segmentation-using-yolov8-neural-network-3if9#prepare_data. Examples in it based on local Jupyter Notebooks. Using Google Colab is out of scope of this article, but it's almost the same. You just need to upload the dataset to your Colab or Google Drive account and then pass the path to it to the Colab notebook. There are a lot of information on how to use Google Colab to work with custom datasets. For example, see this video to learn how to upload and load data from custom dataset for YOLOv8: kzbin.infobx52WmQvbaE?app=desktop, or here is an official Roboflow docs for Google Colab intergration as well: roboflow.com/integration/google-colab

I use Roboflow Annotate to create datasets, annotate them and then train the algorithms on my own. The Roboflow Universe can be used to download already annotated datasets of images, that can be used to train algorithms. Also, the Roboflow can be used to train the model, based on dataset, selected from the Roboflow Universe using their cloud GPUs. But as far, as I know, this feature is not available for free. Also, not sure, that they have collections of already trained models. However, I think you better ask the Roboflow support about available options.

@@germanovdev6964 How can I contact with Roboflow support and ask them ?

Hello. You can use the following code to download the N model and convert it to ONNX: from ultralytics import YOLO model = YOLO("yolov8n-seg.pt") model.export(format="onnx") or the following code to download the S model and convert it to ONNX: from ultralytics import YOLO model = YOLO("yolov8s-seg.pt") model.export(format="onnx")

@@germanovdev6964 Yes. Thank you so much .

Sure, here it is: drive.google.com/file/d/1yeSlei33hGgN0wEeBCHOfcah49CaGnD0/view?usp=sharing. But this sample dataset has only 6 images. You can find more interesting datasets in Roboflow Universe: universe.roboflow.com/search?q=cat+dogs+detection+dataset+model%3Ayolov8+object+detection and download in the YOLOv8 format.

@@germanovdev6964 Thanks

Yes, you can connect a webcam as an input for the <video> element. This article describes how: dev.to/dalalrohit/live-stream-your-webcam-to-html-page-3ehf.

Do you mean YOLOv8 classification model? If yes, then it's true, it pretrained on ImageNet dataset with 1000 classes, that include different breeds of dogs and cats. If you need to predict only dogs and cats, use an object detection model, that pretrained on COCO dataset with 80 classes. Also, you can train your own model for classification, using dogs and cats dataset.

Thank you, please read this article to learn YOLOv8 segmentation step by step: dev.to/andreygermanov/how-to-implement-instance-segmentation-using-yolov8-neural-network-3if9. It also has videos.

To do this, you need to join all detected masks to a single one, pixel by pixel. Let's assume that you predicted the following result: results = model.predict("cat_dog.jpg") result = results[0] First, let's get two masks from this result: mask1 = result.masks.data[0].numpy() mask2 = result.masks.data[1].numpy() By default, the type of pixel values is float32, and it's better to convert them to 'uint8': mask1 = mask1.astype('uint8') mask2 = mask2.astype('uint8') Now, if pixel is black, it has 0 value and if pixel is white it has 1 value. Then you can join these masks by applying bitwise OR operation: mask = mask1 | mask2 It applies OR operation for each pixel of both masks in the following way: 0 | 0 = 0 0 | 1 = 1 1 | 0 = 1 1 | 1 = 1 Finally, the pixels that are white either on the first or on the second mask are written as white to the final mask. Then, you can convert this mask to image and save: from PIL import Image mask = mask * 255 # white color in image should be 255, not 1 img = Image.fromarray(mask, "L") img.save("image.jpg")

I could not reproduce this, but in some system configurations Jupyter do not display images. One guy experienced this before. You can find tips that you can try to do in the comments below. If all this does not work, you can just save this image to a file this way: Image.fromarray(result.plot( )[:,:,::-1]).save("image.jpg") and then open the "image.jpg" by external program.

@@germanovdev6964 Thanks, everything works. I have an idea for another video - face detection from image and head rotation analysis. I'm interested in this and I don't see an off-the-shelf solution based on YOLO. I think it would add views to the channel. ))

Never experienced this issue in VS Code, assuming that the 'result' contains correct image array... You can try two options: 1) Display the image using: from IPython.display import display display(Image.fromarray(result.plot()[:,:,::-1])) 2) Perhaps you've put a semicolon at the end of the line? Do not do this.

This is the notebook: colab.research.google.com/drive/1TtnZXbxqJROHtBFmIZ8jW3_zVP4bzQnI?usp=sharing, but dataset with YAML file is not there, because it is on my Google Drive. Here is the dataset with data.yaml file: drive.google.com/file/d/1PNktsghBqIJVgxa-34FqO3yODNJbH3B0 The whole workflow on this video described in this article: dev.to/andreygermanov/a-practical-introduction-to-object-detection-with-yolov8-neural-network-3n8c#train