DDRNet.TensorRT icon indicating copy to clipboard operation
DDRNet.TensorRT copied to clipboard

TensorRT of DDRNet for real-time segmentation

Results 8 DDRNet.TensorRT issues
Sort by recently updated
recently updated
newest added

如果使用ddrnet.pytorch里面,转成onnx, 再用trtexec 转换成engine文件,推理就可以吧?

想提升输入-输出的总体时间

![image](https://user-images.githubusercontent.com/43496548/170947230-8c2e8433-decd-4bce-a599-5b16640ede33.png)

Hi! The getwts.py script is unusable due to the required modules are not included in the project. Could you upload them or just the DDRNet_CS.wts file to google drive? Thanks!

是不是模型对int8 不支持,如果想跑出int8的速度 需要改训练时的模型,使其最后得到的参数已经是量化的。 然后这个.tensorRT的作用。。。就 是不是和 ddrnet 直接导出 .onnx,然后用官方自带的工具转成trt 效果是一样的

I was wondering if this could support ddrnet23 (not slim) as well

感谢大神提供的代码,受益匪浅! 关于上采样我有个疑问,想恢复准确率的话需要finetune with nearest是指将训练模型的上采样方式改成nearest吗?然后有没有可能在tensorrt的c++代码内修改上采样方式为bilinear呢