Does TensorRt support flexible sizes of input ?

I am using tensor rt to run caffe models.

After I build the engine, must I input the same size of input as the prototxt describes?

I mean , does tensor rt support reshape the input sizes of each layer without rebuilding the engine ?

thank you !


TensorRT does not have adaptive input sizes yet. The user needs to specify their input size at build time in order for us to optimize the network.

This is a frequently requested feature and we taking feedback and planning accordingly.

NVIDIA Enterprise Support