This will depend on the GPU capacity of your PC. Using a PC with a 4G RAM GPU will be 2fps to perform semantic segmentation with a pretrained ade20k model( a model that supports 150 classes of objects). You will achieve a higher inference speed with a GPU of high capacity.