Table 1 Comparison between different loss functions and NWDLOSS.
From: A detection method for small casting defects based on bidirectional feature extraction
CIOU | |
|---|---|
Weaknesses | Cannot optimize when there is no overlap |
Advantages of NWDLOSS | More robust, handles non-overlapping cases and adapts to small object detection |
GIoU | |
Weaknesses | Insensitive to subtle changes |
Advantages of NWDLOSS | Performs better in detail capturing and multi-object scenarios |
DIoU | |
Weaknesses | Ignores aspect ratio influence |
Advantages of NWDLOSS | Considers both position and shape differences, excelling with complex targets |
CIoU | |
Weaknesses | Not sensitive enough to small objects |
Advantages of NWDLOSS | Better suited for small object detection, especially in complex background scenarios |
Smooth L1 Loss | |
Weaknesses | Lacks consideration of box overlap or shape |
Advantages of NWDLOSS | Quantifies box differences with Wasserstein distance, ideal for high-precision tasks |
