can caffe Tile Layer use the your TensorRT plugin?

[shijieliu]

hi, it's awesome to see this project and i am wondering if i can use this in my own code. What i encountered is when i element wise mutiply two matrix whose shape is (2048, 128, 24, 24) and (2018, 128, 1, 1), it add a Tile Layer in prototxt, and it seems like this layer is not supported by the tensorrt 5.0. So i would like to know if i can use your Brodcast code and if i want to use what should i change? thanks.

[junqiangwu]

The broadcast layer requires that the first dimension must be the same, but in your project is 2048 and 2018, I don't know if the Tile Layer can solve this problem. I think you can refer to Caffe's Tile Layer implementation and implement it with Tensor RT, and Tensor RT is very convenient to add custom layers.

[shijieliu]

The broadcast layer requires that the first dimension must be the same, but in your project is 2048 and 2018, I don't know if the Tile Layer can solve this problem. I think you can refer to Caffe's Tile Layer implementation and implement it with Tensor RT, and Tensor RT is very convenient to add custom layers.

Thank you! it is a typo error to mistake 2048 to 2018.

[junqiangwu]

Oh, then you can use the broadcast layer directly, I’m happy to help you.