unable to compile tensorflow mobilenet ssd coco model from TF object detection app zoo using ncsdk2

[idata]

I have installed the latest version of ncsdk2 which provides support for tf ssd as per release notes(2.10) but when I am trying to load the graph file it say dynamic input mode is not supported. The input node name given is image_tensor. Appreciate any support on this issue .. thank you

[idata]

I also encountered the same error.

After run mvNCCompile -h, I found there's a --tf-ssd-config option. So I used the following command to compile the model:

mvNCCompile frozen_inference_graph.pb -s 12 -in image_tensor -on detection_boxes --tf-ssd-config pipeline.config

Now the error message is:

Traceback (most recent call last): File "/usr/local/bin/mvNCCompile", line 208, in <module> args.old_parser, args.cpp, args) File "/usr/local/bin/mvNCCompile", line 186, in create_graph load_ret = load_network(args, parser, myriad_config) File "/usr/local/bin/ncsdk/Controllers/Scheduler.py", line 70, in load_network p.load_ssd_config(arguments.tf_ssd_config) File "/usr/local/bin/ncsdk/Controllers/Parsers/TensorFlow.py", line 237, in load_ssd_config text_format.Parse(config_pb.read(), self.ssd_config) File "/usr/lib/python3.5/site-packages/google/protobuf/text_format.py", line 499, in Parse descriptor_pool=descriptor_pool) File "/usr/lib/python3.5/site-packages/google/protobuf/text_format.py", line 563, in ParseLines return parser.ParseLines(lines, message) File "/usr/lib/python3.5/site-packages/google/protobuf/text_format.py", line 613, in ParseLines self._ParseOrMerge(lines, message) File "/usr/lib/python3.5/site-packages/google/protobuf/text_format.py", line 638, in _ParseOrMerge self._MergeField(tokenizer, message) File "/usr/lib/python3.5/site-packages/google/protobuf/text_format.py", line 730, in _MergeField (message_descriptor.full_name, name)) google.protobuf.text_format.ParseError: 1:1 : Message type "TFSSDConfig" has no field named "model".

And I have no clue about this…

[idata]

I tried to inspect the error message of Message type "TFSSDConfig" has no field named "model"., but still have no solution to it.

The "model" name was from the beginning of the file pipeline.config for the --tf-ssd-config option. The starting lines of my pipeline.config is:

model { ssd { num_classes: 90 image_resizer { fixed_shape_resizer { height: 300 width: 300 } }

If anyone know how to solve it, or know any example of compiling Tensoflow Mobilenet SSD into NCS graph, please give some help. Thank you.

[idata]

I also have same problem.

OS : ubuntu 16.0.4NCSDK : 2.10.01.01Command line : mvNCCompile frozen_inference_graph.pb -s 12 --tf-ssd-config pipeline.config

[idata]

Any updates? Has the same issue here.

[idata]

Hi @eric_huang @Edjoz @hiankun @sreerag_ibtl

We just posted a thread about the new --tf-ssd-config option for the NCSDK v2.10.

Take a look: NCSDK 2.10 OPTION: --tf-ssd-config

Regards,

Jesus

[idata]

Thank you @Jesus_at_Intel. I have some doubts. I couldn't find the parameters 'background_label_id' and 'box_params' in the original pipeline.config. The model was trained using the object detection api.

[idata]

Hi @sreerag_ibtl

The background_label_id is technically a class but it’s used when nothing is detected and will consider it as part of the background. Some networks utilize this background class and other don’t. I believe they are inherently based on how the model is trained.

The box_params values represent the variance of the bounding boxes, the 0.1 value is used for the center of the bounding box and the 0.2 value is used for the width and height settings.

Hope this helps!

Regards,

Jesus