摘要
OrthoNet,一种基于正交滤波器的通道注意力机制。该机制解决了FcaNet中频率选择的问题,并假设了正交性是DCT内核有效性的主要驱动力。通过将该机制集成到ResNet中,并与FcaNet和其他注意力机制进行比较,OrthoNet在Birds、MS-COCO、Places356等数据集上表现出色。在ImageNet数据集上与当前最先进的方法竞争或超越了它们。
本文使用OrthoNet改进YoloV8,在我自己的数据集上实现有效涨点!
文章和代码链接:
https://blog.csdn.net/m0_47867638/article/details/134454223?spm=1001.2014.3001.5502
YOLOv8l summary (fused): 268 layers, 43631280 parameters, 0 gradients, 165.0 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 29/29 [
all 230 1412 0.922 0.957 0.986 0.737
c17 230 131 0.973 0.992 0.995 0.825
c5 230 68 0.945 1 0.995 0.836
helicopter 230 43 0.96 0.907 0.951 0.607
c130 230 85 0.984 1 0.995 0.655
f16 230 57 0.955 0.965 0.985 0.669
b2 230 2 0.704 1 0.995 0.722
other 230 86 0.903 0.942 0.963 0.534
b52 230 70 0.96 0.971 0.978 0.831
kc10 230 62 0.999 0.984 0.99 0.847
command 230 40 0.97 1 0.995 0.811
f15 230 123 0.891 1 0.992 0.701
kc135 230 91 0.971 0.989 0.986 0.712
a10 230 27 1 0.555 0.899 0.456
b1 230 20 0.972 1 0.995 0.793
aew 230 25 0.945 1 0.99 0.784
f22 230 17 0.913 1 0.995 0.725
p3 230 105 0.99 1 0.995 0.801
p8 230 1 0.637 1 0.995 0.597
f35 230 32 0.939 0.938 0.978 0.574
f18 230 125 0.985 0.992 0.987 0.817
v22 230 41 0.983 1 0.995 0.69
su-27 230 31 0.925 1 0.995 0.859
il-38 230 27 0.972 1 0.995 0.811
tu-134 230 1 0.663 1 0.995 0.895
su-33 230 2 1 0.611 0.995 0.796
an-70 230 2 0.766 1 0.995 0.73
tu-22 230 98 0.984 1 0.995 0.831
Speed: 0.2ms preprocess, 3.8ms inference, 0.0ms loss, 0.8ms postprocess per image
YOLOv8l summary (fused): 296 layers, 43934384 parameters, 0 gradients, 165.2 GFLOPs
Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 15/15 [00:03<00:00, 4.96it/s]
all 230 1412 0.974 0.98 0.992 0.758
c17 230 131 0.988 0.992 0.995 0.839
c5 230 68 0.963 1 0.995 0.841
helicopter 230 43 0.974 1 0.986 0.61
c130 230 85 1 0.977 0.995 0.673
f16 230 57 0.984 0.965 0.989 0.689
b2 230 2 0.907 1 0.995 0.847
other 230 86 0.942 0.965 0.973 0.564
b52 230 70 0.981 0.971 0.987 0.844
kc10 230 62 1 0.984 0.989 0.836
command 230 40 0.99 1 0.995 0.848
f15 230 123 0.997 0.976 0.995 0.695
kc135 230 91 0.987 0.989 0.99 0.693
a10 230 27 1 0.719 0.97 0.461
b1 230 20 1 0.965 0.995 0.742
aew 230 25 0.951 1 0.992 0.784
f22 230 17 0.982 1 0.995 0.771
p3 230 105 1 0.966 0.995 0.8
p8 230 1 0.859 1 0.995 0.796
f35 230 32 0.998 1 0.995 0.583
f18 230 125 0.984 0.992 0.992 0.83
v22 230 41 0.994 1 0.995 0.69
su-27 230 31 0.986 1 0.995 0.858
il-38 230 27 0.989 1 0.995 0.866
tu-134 230 1 0.847 1 0.995 0.895
su-33 230 2 0.987 1 0.995 0.697
an-70 230 2 1 1 0.995 0.858
tu-22 230 98 0.996 1 0.995 0.843
Speed: 0.2ms preprocess, 9.6ms inference, 0.0ms loss, 0.8ms postprocess per image
提升的幅度还不小!