* separate model and block * move thresh to config * use trt8 api * move batchsize to config * update readme * refactor batch infer logic * move max input image size to config