去年11月份，谷歌大脑提出兼顾准确率和模型效率的新型目标检测器EfficietDet，实现了新的SOTA结果。
Yet-Aother-EfficietDet-Pytorch 是具有SOTA实时性能的官方 EfficietDet 的pytorch重新实现。该版本的性能接近原版，但速度是官方TesorFlow实现的近26倍！
性能Pretraiedweightsadbechmark
coefficietpth_dowloadGPUMem(MB)FPSExtremeFPS(Batchsize32)mAP0.5:0.95(thisrepo)mAP0.5:0.95(paper)D0efficietdet-d0.pth104936.20163.1432.633.8D1efficietdet-d1.pth115929.6953.8238.239.6D2efficietdet-d2.pth132126.5040.4341.543.0D3efficietdet-d3.pth164722.73-44.945.8D4efficietdet-d4.pth190314.75-48.149.4D5efficietdet-d5.pth22557.11-49.550.7D6efficietdet-d6.pth29855.30-50.151.7D7efficietdet-d7.pth38193.73-50.752.2SpeedTest
coefficietTimeFPSRatioOfficialD0(tfpostprocess)0.713s1.401XOfficialD0(umpypostprocess)0.477s2.091.49XYet-Aother-EfficietDet-D00.028s36.2025.86X
评论