profile = builder.create_optimization_profile()
profile.set_shape(network.get_input(0).name, (1,3, 32, 32), (1,3, 608, 608), (1,3, 2050, 2050))
config.add_optimization_profile(profile)

config.max_workspace_size=1<<30 #1GB

config.max_workspace_size

engine = builder.build_engine(network,config)

inputs = []
outputs = []
bindings = []
stream = cuda.Stream()
tmp=[1,32,16,8]
print('engine.get_binding_format_desc',engine.get_binding_format_desc(0))
for count,binding in enumerate(engine):print('binding:',binding)size = trt.volume(engine.get_binding_shape(binding)) * engine.max_batch_size*(int)(h_/tmp[count])*(int)(w_/tmp[count])  dtype = trt.nptype(engine.get_binding_dtype(binding))  # Allocate host and device buffershost_mem = cuda.pagelocked_empty(size, dtype)device_mem = cuda.mem_alloc(host_mem.nbytes)# Append the device buffer to device bindings.bindings.append(int(device_mem))# Append to the appropriate list.if engine.binding_is_input(binding):inputs.append(HostDeviceMem(host_mem, device_mem))else:outputs.append(HostDeviceMem(host_mem, device_mem))

trt_outputs = common.do_inference_v3(self.context,bindings=self.bindings,inputs=self.inputs,outputs=self.outputs,stream=self.stream,h_=h_,w_=w_)output_shapes = [(1, int(h_/32), int(w_/32), 21),(1, int(h_/16), int(w_/16), 21),(1, int(h_/8), int(w_/8), 21)]print(output_shapes)
trt_outputs = [output[:shape[1]*shape[2]*shape[3]].reshape(shape) for output, shape in zip(trt_outputs, output_shapes)]