How to deduce the output_node_names of a pretrained model?

Question

I'm trying to convert a pretrained model for us with Tensorflow.js:

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

I chose mask_rcnn_inception_v2_coco.

tensorflowjs_converter expects specific output_node_names. Various resources on the web point me to tools like summarize_graph to help with inspecting potential output node names.

Unfortunately, I'm running this on Google Colab, and (from what I can tell) I can't install bazel, which I need to build summarize_graph from source, which I need to identify which output_node_names to pass to the converter.

Am I missing something here? Is there a more straight forward way to go from an existing pretrained model to Tensorflow.js (for inference on the browser)?

Answer 1

For mask_rcnn_inception_v2_coco_2018_01_28 the result of bazel-bin/tensorflow/tools/graph_transforms/summarize_graph --in_graph=frozen_inference_graph.pb is

Found 1 possible inputs: (name=image_tensor, type=uint8(4), shape=[?,?,?,3])
No variables spotted.
Found 5 possible outputs: (name=detection_boxes, op=Identity) (name=detection_scores, op=Identity) (name=detection_classes, op=Identity) (name=num_detections, op=Identity) (name=detection_masks, op=Identity) 

Answer 2

This article helped me: https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/optimizing-neural-networks-for-mobile-and-embedded-devices-with-tensorflow/determine-the-names-of-input-and-output-nodes

Get tensorboard up and running (can be done in windows or whatever, this is an ubuntu install)

pip install tensorboard

>

tensorboard --logdir=/tmp/tensorboard

>

python tensorflow/python/tools/import_pb_to_tensorboard.py --model_dir resnetv1_50.pb --log_dir /tmp/tensorboard

(if you don't have import_pb_to_tensorboard.py you can just download file https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/import_pb_to_tensorboard.py and point to on whatever directory you save it)

You use the above command to import you model into the tensorboard you are serving. In tensorboard you can click into the model and see what the final output is called. Mine is called "import/final_result"