retinaface icon indicating copy to clipboard operation
retinaface copied to clipboard

Published 20 hours ago • verified publisher icon clancylian

tensort模型检测结果与caffe模型相差较大

Open oxygenman opened this issue 4 years ago • 1 comments

同样的是mnet-deconv-0517模型,输入图片大小为(1200,660),阈值为0.9的情况下,caffe模型的检测结果是: image 而tensorrt的模型却是这样的? image 差别为啥这么大呢?

oxygenman avatar Sep 01 '20 03:09 oxygenman

您好，我同样使用TensorRT去跑demo，效果也比较差，6个人只能检测出4个。当我使用caffe时，却出现 Check failed: status == CUDNN_STATUS_SUCCESS (4 vs. 0) CUDNN_STATUS_INTERNAL_ERROR，请问这个怎么解决呢？ ps：使用cuda10.1和cudnn7.6.5

crazybill-first avatar Jan 29 '21 07:01 crazybill-first