YOLOv8 latest improvement series: Optimizing CIoU to Wise-IoU
Click here for the address of Wise-IoU's paper. It complements the theory and is far ahead.
Get the improved source code here o
As of press time, the source code package of the latest improvement series of YOLOv8 at Bilibili has been updated with 22 network improvements + Wise-IoU loss function improvements!
YOLOv8’s latest improvement series: Optimize CIoU to Wise-IoU to help improve detection performance and effectively improve small target detection accuracy!
1. Overview of Wise-IoU
Original abstract: Target detection is a core issue in computer vision, and its detection performance depends on the design of the loss function. The bounding box loss function is an important part of the target detection loss function, and its good definition will bring significant performance improvements to the target detection model. Most of the research in recent years assumes that the examples in the training data are of high quality and focuses on strengthening the fitting ability of the bounding box loss. However, we noticed that the target detection training set contains low-quality examples. If we blindly strengthen the regression of bounding boxes on low-quality examples, it will obviously harm the improvement of model detection performance. Focal-EIoU v1 was proposed to solve this problem, but because its focusing mechanism is static, the potential of the non-monotonic focusing mechanism is not fully exploited. Based on this point of view, we proposed a dynamic non-monotonic focusing mechanism and designed Wise-IoU (WIoU). The dynamic non-monotonic focusing mechanism uses "outlier" instead of IoU to evaluate the quality of anchor boxes and provides a wise gradient gain allocation strategy. This strategy reduces the competitiveness of high-quality anchor boxes while also reducing the harmful gradients produced by low-quality examples. This allows WIoU to focus on normal quality anchor boxes and improve the overall performance of the detector. When WIoU is applied to the state-of-the-art single-stage detector YOLOv7, the AP-75 on the MS-COCO dataset increases from 53.03% to 54.50%
2. YOLOv8 latest improvement series: CIoU changed to Wise-IoU method
All improved codes have been placed in the blogger's workshop!
Link here!
Poke him!
3. The final verification is successful.
Excuting an order
python train.py
It’s ok if you run through it!
Changes completed and call it a day!
Pay attention to Station B: AI academic is called a beast,
and it has embarked on the fast path of scientific research,
far ahead of its peers! ! ! !