/Stronger-yolo

šŸ”„Improve yolo with latest paper

Primary LanguagePythonMIT LicenseMIT

train dataset: VOC 2012 + VOC 2007
test size: 544
test code: Faster rcnn (not use 07 metric)
test GPU: 12G 2080Ti
test CPU: E5-2678 v3 @ 2.50GHz

VersionNetworkBackboneInitial weightVOC2007 Test(mAP)Inference(GPU)Inference(CPU)Params
V1YOLOV3Darknet53YOLOV3-608.weights88.830.0ms255.8ms248M
V2YOLOV3Darknet53Darknet53_448.weights83.330.0ms255.8ms248M
V3YOLOV3-LiteMobilenetV2MobilenetV2_1.0_224.ckpt79.418.9ms80.9ms27.3M

Check Strongeryolo-pytorch for pytorch version with channel-pruning.
There is also a MNN Demo for Verson V3.