tflight graph created wrongly

和自甴很熟 提交于 2019-12-13 03:09:10

问题


I have the following frozen inference graph. This is for semantic segmentation using Deeplab (download graph here). I converting this graph to tflite format

tflite_convert \
  --output_file=test2.lite \
  --graph_def_file=frozen_inference_graph_3mbvoc.pb \
  --input_arrays=ImageTensor \
  --output_arrays=SemanticPredictions \
  --input_shapes=1,450,600,3 \
  --inference_input_type=QUANTIZED_UINT8 \
  --inference_type=FLOAT \
  --mean_values=128 \
  --std_dev_values=128

After conversion the graph looks as follows(download it here) My question is how do I obtain graph similar to googles graph of deeplab available (here) ? To give you a more clearer question please see below image

the graph on left is my tflite graph and the graph on right is graph of deeplab by google. How do I obtain results similar to graph on right?


回答1:


tflite_convert \
  --output_file=test2.lite \
  --graph_def_file=frozen_inference_graph_3mbvoc.pb \
  --input_arrays=sub_2 \
  --output_arrays=ResizeBilinear_2 \
  --input_shapes=1,450,600,3 \
  --inference_input_type=QUANTIZED_UINT8 \
  --inference_type=FLOAT \
  --mean_values=128 \
  --std_dev_values=128

This solved my question



来源:https://stackoverflow.com/questions/55808705/tflight-graph-created-wrongly

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!