I’m doing a project about image segmentation. I have trained a Mask-RCNN model using tensorflow object detection API however the inference time is quite long (about 2s per frame on my CPU), I have use a Openvino (an Intel framework to optimize DL model for embedded system) but the result is not improve much.
Anyone have experience on it can shed me a light on which model should I use to infer as fast as possible. Deeplab can get better result ? I know it is not instance segmentation but if it is really fast I can give it a try. And how about inference time of Unet ?
Thank you in advance,