Dynamic image resolution in fully convolutional networks?

Question

planning to convert a fully convolutional onnx model into TensorRT. Would like to make this model dynamic to differing image resolutions i.e. 1920x1080, 1280x720, 640x480 et