weixin_39930671
weixin_39930671
2020-12-01 14:40

Questions about using XportDNN

I was trying using XportDNN to convert my prototxt of yolov3 to quantization mode. But there is an error when i run the script like this: aaa1213 However, i have added the upsample layer to my caffe. So i wonder whether the XportDNN can convert the layer added by users or not. And i also want to know how you handle this issue (about the quantization params of Upsample layer & Yolo layer) cause i saw the Benchmarks at Performance_eval.md benchmarks

该提问来源于开源项目:Xilinx/CHaiDNN

  • 点赞
  • 写回答
  • 关注问题
  • 收藏
  • 复制链接分享
  • 邀请回答

5条回答

  • weixin_39824898 weixin_39824898 5月前

    CHaiDNN does not support leaky ReLU. So you need to retrain all the Yolo networks in darknet with normal ReLU. But if you retrain with normal ReLU, it is very difficult to converge.

    点赞 评论 复制链接分享
  • weixin_39930671 weixin_39930671 5月前

    thanks for your hint, i'll figure out this later. Do you know the answer of what i mentioned about the XportDNN?

    点赞 评论 复制链接分享
  • weixin_39824898 weixin_39824898 5月前

    https://github.com/Xilinx/CHaiDNN/issues/124

    点赞 评论 复制链接分享
  • weixin_39620845 weixin_39620845 5月前

    I do the same work as you,how do you solve your question?how does yolov3 work successful????

    点赞 评论 复制链接分享
  • weixin_39930671 weixin_39930671 5月前

    Not yet. If you make any progress, please tell me. Thanks

    点赞 评论 复制链接分享

相关推荐