Why the detection results of consecutive similar images are different？

[daigang896]

Search before asking

• [X] I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

Hello. Why the detection results of consecutive similar images are different？For example, for the same license plate, two consecutive similar license plates, and the same model, through model prediction, one has a target box, and the other has no target box?

Additional

No response

[AI-Expert-04]

yolov5 has a lot of optimizations for performance. In deep learning, increasing accuracy decreases speed, and increasing speed decreases accuracy. In addition, the resolution of the image to be learned, Accuracy depends on how you classify train_set, valid_set, and test_set into ratios.

Especially in image processing, the predicted value depends on the resolution of the video or image to be detected. Even really similar images may or may not be predicted because YOLO is a multi-detection model.

There are ways to increase both performance and accuracy.

1. This is a transfer learning method using the YOLO model as weights.
2. Import enough data set before transfer learning with YOLO model.
3. The higher the resolution of the image, the better.
4. Train_set(70%) valid_set(25%) test_set(5%) Classify the data.
5. The higher the number of lessons, the better.

If you meet all of the above, you can take advantage of both the incredible speed and accuracy of YOLO.

[daigang896]

Hello. It is not to improve the accuracy, but not to understand that the detection results obtained from similar images are different. I want to have a deeper understanding of the algorithm itself and see if there is a good solution.

[github-actions[bot]]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!

[glenn-jocher]

@daigang896 hi there,

Thank you for your interest in understanding the YOLOv5 algorithm! The variation in detection results for similar consecutive images could be attributed to several factors such as object occlusion, lighting conditions, or image noise, which influence the model's predictions. YOLOv5 leverages a deep learning approach that analyzes images holistically and may interpret similar scenes differently due to minor variations.

One potential solution to address this variability is to aggregate multiple detections over time to increase robustness, or to explore ensemble methods that combine outputs from multiple models for improved stability.

Understanding the algorithm's behavior and fine-tuning the model's input and architecture parameters can further enhance its performance. Feel free to refer to the Ultralytics Docs for detailed information on model configuration and training techniques.

Your insight into achieving a balance between speed and accuracy with YOLOv5 is appreciated, and the YOLO community and the Ultralytics team continuously work to optimize its capabilities.