Pytorch YOLOv3 网络结构

缺乏、安全感 2022-04-06 02:48 329阅读 0赞

*  github：[https://github.com/talebolano/yolov3-network-slimming][https_github.com_talebolano_yolov3-network-slimming]

Pytorch YOLOv3 结构  
ModuleList(  
(0): Sequential(  
(conv\_with\_bn\_0): Conv2d(3, 32, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_0): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_0): ReLU(inplace)  
)  
(1): Sequential(  
(conv\_with\_bn\_1): Conv2d(32, 64, kernel\_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)  
(batch\_norm\_1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_1): ReLU(inplace)  
)  
(2): Sequential(  
(conv\_with\_bn\_2): Conv2d(64, 32, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_2): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_2): ReLU(inplace)  
)  
(3): Sequential(  
(conv\_with\_bn\_3): Conv2d(32, 64, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_3): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_3): ReLU(inplace)  
)  
(4): Sequential(  
(shortcut\_4): shortcutLayer()  
)  
(5): Sequential(  
(conv\_with\_bn\_5): Conv2d(64, 128, kernel\_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)  
(batch\_norm\_5): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_5): ReLU(inplace)  
)  
(6): Sequential(  
(conv\_with\_bn\_6): Conv2d(128, 64, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_6): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_6): ReLU(inplace)  
)  
(7): Sequential(  
(conv\_with\_bn\_7): Conv2d(64, 128, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_7): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_7): ReLU(inplace)  
)  
(8): Sequential(  
(shortcut\_8): shortcutLayer()  
)  
(9): Sequential(  
(conv\_with\_bn\_9): Conv2d(128, 64, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_9): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_9): ReLU(inplace)  
)  
(10): Sequential(  
(conv\_with\_bn\_10): Conv2d(64, 128, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_10): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_10): ReLU(inplace)  
)  
(11): Sequential(  
(shortcut\_11): shortcutLayer()  
)  
(12): Sequential(  
(conv\_with\_bn\_12): Conv2d(128, 256, kernel\_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)  
(batch\_norm\_12): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_12): ReLU(inplace)  
)  
(13): Sequential(  
(conv\_with\_bn\_13): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_13): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_13): ReLU(inplace)  
)  
(14): Sequential(  
(conv\_with\_bn\_14): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_14): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_14): ReLU(inplace)  
)  
(15): Sequential(  
(shortcut\_15): shortcutLayer()  
)  
(16): Sequential(  
(conv\_with\_bn\_16): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_16): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_16): ReLU(inplace)  
)  
(17): Sequential(  
(conv\_with\_bn\_17): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_17): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_17): ReLU(inplace)  
)  
(18): Sequential(  
(shortcut\_18): shortcutLayer()  
)  
(19): Sequential(  
(conv\_with\_bn\_19): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_19): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_19): ReLU(inplace)  
)  
(20): Sequential(  
(conv\_with\_bn\_20): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_20): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_20): ReLU(inplace)  
)  
(21): Sequential(  
(shortcut\_21): shortcutLayer()  
)  
(22): Sequential(  
(conv\_with\_bn\_22): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_22): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_22): ReLU(inplace)  
)  
(23): Sequential(  
(conv\_with\_bn\_23): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_23): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_23): ReLU(inplace)  
)  
(24): Sequential(  
(shortcut\_24): shortcutLayer()  
)  
(25): Sequential(  
(conv\_with\_bn\_25): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_25): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_25): ReLU(inplace)  
)  
(26): Sequential(  
(conv\_with\_bn\_26): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_26): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_26): ReLU(inplace)  
)  
(27): Sequential(  
(shortcut\_27): shortcutLayer()  
)  
(28): Sequential(  
(conv\_with\_bn\_28): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_28): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_28): ReLU(inplace)  
)  
(29): Sequential(  
(conv\_with\_bn\_29): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_29): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_29): ReLU(inplace)  
)  
(30): Sequential(  
(shortcut\_30): shortcutLayer()  
)  
(31): Sequential(  
(conv\_with\_bn\_31): Conv2d(256, 512, kernel\_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)  
(batch\_norm\_31): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_31): ReLU(inplace)  
)  
(32): Sequential(  
(conv\_with\_bn\_32): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_32): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_32): ReLU(inplace)  
)  
(33): Sequential(  
(conv\_with\_bn\_33): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_33): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_33): ReLU(inplace)  
)  
(34): Sequential(  
(shortcut\_34): shortcutLayer()  
)  
(35): Sequential(  
(conv\_with\_bn\_35): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_35): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_35): ReLU(inplace)  
)  
(36): Sequential(  
(conv\_with\_bn\_36): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_36): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_36): ReLU(inplace)  
)  
(37): Sequential(  
(shortcut\_37): shortcutLayer()  
)  
(38): Sequential(  
(conv\_with\_bn\_38): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_38): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_38): ReLU(inplace)  
)  
(39): Sequential(  
(conv\_with\_bn\_39): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_39): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_39): ReLU(inplace)  
)  
(40): Sequential(  
(shortcut\_40): shortcutLayer()  
)  
(41): Sequential(  
(conv\_with\_bn\_41): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_41): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_41): ReLU(inplace)  
)  
(42): Sequential(  
(conv\_with\_bn\_42): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_42): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_42): ReLU(inplace)  
)  
(43): Sequential(  
(shortcut\_43): shortcutLayer()  
)  
(44): Sequential(  
(conv\_with\_bn\_44): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_44): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_44): ReLU(inplace)  
)  
(45): Sequential(  
(conv\_with\_bn\_45): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_45): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_45): ReLU(inplace)  
)  
(46): Sequential(  
(shortcut\_46): shortcutLayer()  
)  
(47): Sequential(  
(conv\_with\_bn\_47): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_47): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_47): ReLU(inplace)  
)  
(48): Sequential(  
(conv\_with\_bn\_48): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_48): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_48): ReLU(inplace)  
)  
(49): Sequential(  
(shortcut\_49): shortcutLayer()  
)  
(50): Sequential(  
(conv\_with\_bn\_50): Conv2d(512, 1024, kernel\_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)  
(batch\_norm\_50): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_50): ReLU(inplace)  
)  
(51): Sequential(  
(conv\_with\_bn\_51): Conv2d(1024, 512, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_51): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_51): ReLU(inplace)  
)  
(52): Sequential(  
(conv\_with\_bn\_52): Conv2d(512, 1024, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_52): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_52): ReLU(inplace)  
)  
(53): Sequential(  
(shortcut\_53): shortcutLayer()  
)  
(54): Sequential(  
(conv\_with\_bn\_54): Conv2d(1024, 512, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_54): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_54): ReLU(inplace)  
)  
(55): Sequential(  
(conv\_with\_bn\_55): Conv2d(512, 1024, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_55): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_55): ReLU(inplace)  
)  
(56): Sequential(  
(shortcut\_56): shortcutLayer()  
)  
(57): Sequential(  
(conv\_with\_bn\_57): Conv2d(1024, 512, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_57): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_57): ReLU(inplace)  
)  
(58): Sequential(  
(conv\_with\_bn\_58): Conv2d(512, 1024, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_58): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_58): ReLU(inplace)  
)  
(59): Sequential(  
(shortcut\_59): shortcutLayer()  
)  
(60): Sequential(  
(conv\_with\_bn\_60): Conv2d(1024, 512, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_60): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_60): ReLU(inplace)  
)  
(61): Sequential(  
(conv\_with\_bn\_61): Conv2d(512, 1024, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_61): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_61): ReLU(inplace)  
)  
(62): Sequential(  
(conv\_with\_bn\_62): Conv2d(1024, 512, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_62): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_62): ReLU(inplace)  
)  
(63): Sequential(  
(conv\_with\_bn\_63): Conv2d(512, 1024, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_63): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_63): ReLU(inplace)  
)  
(64): Sequential(  
(conv\_without\_bn\_64): Conv2d(1024, 33, kernel\_size=(1, 1), stride=(1, 1))  
)  
(65): Sequential(  
(Detection\_65): DetectionLayer(  
(mse\_loss): MSELoss()  
(bce\_loss): BCELoss()  
(ce\_loss): CrossEntropyLoss()  
)  
)  
(66): Sequential(  
(route\_66): Route()  
)  
(67): Sequential(  
(conv\_with\_bn\_67): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_67): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_67): ReLU(inplace)  
)  
(68): Sequential(  
(upsample\_68): Upsample(scale\_factor=2, mode=nearest)  
)  
(69): Sequential(  
(route\_69): Route()  
)  
(70): Sequential(  
(conv\_with\_bn\_70): Conv2d(768, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_70): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_70): ReLU(inplace)  
)  
(71): Sequential(  
(conv\_with\_bn\_71): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_71): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_71): ReLU(inplace)  
)  
(72): Sequential(  
(conv\_with\_bn\_72): Conv2d(512, 256, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_72): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_72): ReLU(inplace)  
)  
(73): Sequential(  
(conv\_with\_bn\_73): Conv2d(256, 512, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_73): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_73): ReLU(inplace)  
)  
(74): Sequential(  
(conv\_without\_bn\_74): Conv2d(512, 33, kernel\_size=(1, 1), stride=(1, 1))  
)  
(75): Sequential(  
(Detection\_75): DetectionLayer(  
(mse\_loss): MSELoss()  
(bce\_loss): BCELoss()  
(ce\_loss): CrossEntropyLoss()  
)  
)  
(76): Sequential(  
(route\_76): Route()  
)  
(77): Sequential(  
(conv\_with\_bn\_77): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_77): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_77): ReLU(inplace)  
)  
(78): Sequential(  
(upsample\_78): Upsample(scale\_factor=2, mode=nearest)  
)  
(79): Sequential(  
(route\_79): Route()  
)  
(80): Sequential(  
(conv\_with\_bn\_80): Conv2d(384, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_80): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_80): ReLU(inplace)  
)  
(81): Sequential(  
(conv\_with\_bn\_81): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_81): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_81): ReLU(inplace)  
)  
(82): Sequential(  
(conv\_with\_bn\_82): Conv2d(256, 128, kernel\_size=(1, 1), stride=(1, 1), bias=False)  
(batch\_norm\_82): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_82): ReLU(inplace)  
)  
(83): Sequential(  
(conv\_with\_bn\_83): Conv2d(128, 256, kernel\_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)  
(batch\_norm\_83): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track\_running\_stats=True)  
(relu\_83): ReLU(inplace)  
)  
(84): Sequential(  
(conv\_without\_bn\_84): Conv2d(256, 33, kernel\_size=(1, 1), stride=(1, 1))  
)  
(85): Sequential(  
(Detection\_85): DetectionLayer(  
(mse\_loss): MSELoss()  
(bce\_loss): BCELoss()  
(ce\_loss): CrossEntropyLoss()  
)  
)  
)

[https_github.com_talebolano_yolov3-network-slimming]: https://github.com/talebolano/yolov3-network-slimming