deeplabv3似乎argmax输出不支持tensorrt部署?

当output_op是none的时候没有任何问题，换成argmax时加载模型出现以下报错
[graphShapeAnalyzer.cpp::analyzeShapes::1872] Error Code 4: Miscellaneous (IShuffleLayer arg_max (Output: argmax_0.tmp_01928): reshape dimension with more than one -1 wildcard. Reshaping [1,1,(# 2 (SHAPE x)),(# 3 (SHAPE x))] to [1,-1,-1].)
以下是加载模型代码
option=fd.RuntimeOption()
ption.use_gpu(0)
option.use_trt_backend()
option.enable_paddle_to_trt()
option.enable_paddle_trt_collect_shape()
option.set_trt_input_shape("x", [1, 3, 256, 256], [1,3,512,512],[1, 3, 1024, 1024])
option.enable_trt_fp16()
model_path=os.path.join(path,"model.pdmodel")
params_file = os.path.join(path, "model.pdiparams")
config_file = os.path.join(path, "deploy.yaml")
model=fd.vision.segmentation.PaddleSegModel(model_path,params_file,config_file，runtime_option=option)

Start with the PaddleSegModel construction and RuntimeOption TensorRT configuration shown in the report, using model.pdmodel, model.pdiparams, and deploy.yaml as the reproduction inputs. Compare loading with output_op set to none versus argmax and use the graphShapeAnalyzer error as the baseline; done means the argmax model loads successfully with the configured input shapes.