onnx.shape_inference.infer_shapes(SqueezeNet) not complete

I want to get inner node output to analyze onnx graph downloaded in https://github.com/onnx/models/tree/master/squeezenet. However, when I code as follow:

import onnx 
import onnx.shape_inference
onnx_model = onnx.load_model('squeezenet.onnx')
onnx_model = onnx.shape_inference.infer_shapes(onnx_model)
print onnx_model.graph.value_info

I only get the first node output, though the second node is just a relu op.
Could I solve this problem easily? Or I should write a shape_inference by my self ?

Start with the provided Python reproduction using onnx.shape_inference.infer_shapes and the SqueezeNet model from onnx/models. Compare the populated graph.value_info entries with the model's intermediate nodes and determine why inference stops after the first output. Done means the issue's missing inner-node outputs are consistently available after shape inference, with a regression check for SqueezeNet.