About the project HumanDetection_YOLOv8

Table of contents (Оглавление)

1. A little bit about the dataset
2. The result of the trained model - YOLO8 Medium custom
3. The result of the trained model - Fine-tuning YOLO8 Medium custom
4. Conclusion

01. A little bit about the dataset

• This dataset was created by extracting frames from 12 short videos. To enrich this data collection with a variety of images, such as blurry images, small-sized images, and others, it was decided to use the unmarked data set available on Kaggle.

• After the required number of frames was obtained, all images were annotated using the Roboflow platform.

02. The result of the trained model - YOLO8 Medium custom

Training model

Model Validation

Predict on images

Predict on video

03. The result of the trained model - Fine-tuning YOLO8 Medium custom

Training model

Model Validation

Predict on images

Predict on video

04. Conclusion

As you can see, a trained model using fine tuning shows better results. Does not make mistakes when detecting an object, as it happens with a model without fine tuning