Yes, absolutely! YOLO models, including YOLOv11, support transfer learning, allowing you to train only the last layers on your custom dataset instead of starting from scratch. This approach is efficient, as it utilizes pre-trained weights, reducing training time and resource requirements. To achieve this, you can freeze the earlier layers of the model (like the backbone) and only update the final layers during training. For example, with YOLOv5, you can use the `--freeze` argument in the training script to freeze specific layers. For YOLOv11, the process is similar and can be configured in the training settings. Check out our guide on Transfer Learning with Frozen Layers docs.ultralytics.com/yolov5/tutorials/transfer_learning_with_frozen_layers/ for specific steps and commands. 🚀
Yes, you can use your annotated dataset from Roboflow for instance segmentation with Ultralytics YOLO models! Just ensure your dataset is in a compatible format (e.g., COCO or YOLO). You can refer to the COCO-Seg dataset documentation docs.ultralytics.com/datasets/segment/coco/ for guidance on dataset structure or use Roboflow's export feature to format it correctly. Let us know if you encounter any issues! 🚀
Yes, when training with `imgsz` in YOLO11, the model resizes images so their largest dimension matches the specified value (e.g., 640 px), while maintaining the original aspect ratio. This ensures that the objects' proportions remain intact without distortion. For more details, check the training docs docs.ultralytics.com/modes/train/.
That's fantastic! Increasing `imgsz` to 1280 allows the model to process images at a higher resolution, which can significantly improve the detection of very small objects by preserving more detail. It's a useful strategy for datasets where small object accuracy is critical. For more on resizing and preprocessing, check this guide: docs.ultralytics.com/guides/preprocessing_annotated_data/. 🚀
To use your custom dataset in an API for object detection and counting, you can integrate Ultralytics YOLO with a backend API and pass detection results to your React Native app. Here's how: 1. Train your custom dataset following this guide docs.ultralytics.com/datasets/. 2. Deploy the trained YOLO model in an API using tools like Flask or FastAPI. The backend should handle image uploads, run YOLO inference, and return detection results. 3. Use the React Native app to send images to the API, receive object counts, and display results. Check the Ultralytics Docs for API integration docs.ultralytics.com/hub/inference-api/ for guidance on setting up inference. Let us know if you need more help! 🚀
Two tones only? Ultralytics is fantastic, it would be nice if you explained it using more then only 2 tones in your voice. All the video's I was wondering, why you talk so strange, please listen to yourself. You can improve and should improve if you are the main spokesman for ultralytics youtube video's. Good luck.
Thanks for the feedback! 😊 We’re always looking to improve not just content but delivery too. I’ll keep this in mind for future videos to ensure they’re more engaging and dynamic. Your support means a lot-thank you for watching! 🚀
Can I do transfer learning with this yolo models ? like training just last layer not from the scratch on my custom dataset?
Yes, absolutely! YOLO models, including YOLOv11, support transfer learning, allowing you to train only the last layers on your custom dataset instead of starting from scratch. This approach is efficient, as it utilizes pre-trained weights, reducing training time and resource requirements.
To achieve this, you can freeze the earlier layers of the model (like the backbone) and only update the final layers during training. For example, with YOLOv5, you can use the `--freeze` argument in the training script to freeze specific layers. For YOLOv11, the process is similar and can be configured in the training settings.
Check out our guide on Transfer Learning with Frozen Layers docs.ultralytics.com/yolov5/tutorials/transfer_learning_with_frozen_layers/ for specific steps and commands. 🚀
I have created project "instance segmentation" type and annotated 300 images on roboflow , can I use that dataset for this ?
Yes, you can use your annotated dataset from Roboflow for instance segmentation with Ultralytics YOLO models! Just ensure your dataset is in a compatible format (e.g., COCO or YOLO). You can refer to the COCO-Seg dataset documentation docs.ultralytics.com/datasets/segment/coco/ for guidance on dataset structure or use Roboflow's export feature to format it correctly. Let us know if you encounter any issues! 🚀
Training useing imgsz with a different value keeps the aspect ratio of objects and does not disturb the images?
Yes, when training with `imgsz` in YOLO11, the model resizes images so their largest dimension matches the specified value (e.g., 640 px), while maintaining the original aspect ratio. This ensures that the objects' proportions remain intact without distortion. For more details, check the training docs docs.ultralytics.com/modes/train/.
Thanks I have trained with imgz = 1280 and obtained better accuracy in detecting very small objects from the image
That's fantastic! Increasing `imgsz` to 1280 allows the model to process images at a higher resolution, which can significantly improve the detection of very small objects by preserving more detail. It's a useful strategy for datasets where small object accuracy is critical. For more on resizing and preprocessing, check this guide: docs.ultralytics.com/guides/preprocessing_annotated_data/. 🚀
I have followed this with my custom dataset which I want to make in API? i want to use in RN to pass image and count objects detected. How to achieve?
To use your custom dataset in an API for object detection and counting, you can integrate Ultralytics YOLO with a backend API and pass detection results to your React Native app. Here's how:
1. Train your custom dataset following this guide docs.ultralytics.com/datasets/.
2. Deploy the trained YOLO model in an API using tools like Flask or FastAPI. The backend should handle image uploads, run YOLO inference, and return detection results.
3. Use the React Native app to send images to the API, receive object counts, and display results.
Check the Ultralytics Docs for API integration docs.ultralytics.com/hub/inference-api/ for guidance on setting up inference. Let us know if you need more help! 🚀
Two tones only?
Ultralytics is fantastic, it would be nice if you explained it using more then only 2 tones in your voice. All the video's I was wondering, why you talk so strange, please listen to yourself. You can improve and should improve if you are the main spokesman for ultralytics youtube video's. Good luck.
Thanks for the feedback! 😊 We’re always looking to improve not just content but delivery too. I’ll keep this in mind for future videos to ensure they’re more engaging and dynamic. Your support means a lot-thank you for watching! 🚀