output_names=[out.op.name for out in model.outputs][:4] should be modified to get the 5 loss nodes

[cpoptic]

Line 50 of main.py:

def freeze_model(model, name):
    frozen_graph = freeze_session(
        sess,
        output_names=[out.op.name for out in model.outputs][:4])

Correct me if I'm wrong, but using the first 4 output names [:4] of your output_names is not guaranteed to get the correct output node names.

What we really want are the 5 output nodes regarding Mask RCNN loss:

rpn_class_loss
rpn_bbox_loss
mrcnn_class_loss
mrcnn_bbox_loss
mrcnn_mask_loss

When I run [out.op.name for out in keras_model.model.outputs]
I get 14 output nodes:

['rpn_class_logits/concat',
'rpn_class/concat',
'rpn_bbox/concat',
'mrcnn_class_logits/Reshape_1',
'mrcnn_class/Reshape_1',
'mrcnn_bbox/Reshape',
'mrcnn_mask/Reshape_1',
'ROI/packed_2',
'output_rois/mul',
'rpn_class_loss/cond/Merge',
'rpn_bbox_loss/cond/Merge',
'mrcnn_class_loss/truediv',
'mrcnn_bbox_loss/Mean',
'mrcnn_mask_loss/Mean']

So, at least for this particular model, the desired output nodes are actually the last 5 of these nodes: `output_names=[out.op.name for out in model.outputs][-5:])

Am I missing something?

[bendangnuksung]

When I added and run

# main.py  line no: 48
 print([out.op.name for out in model.outputs])

I am getting this values ['mrcnn_detection/Reshape_1', 'mrcnn_class/Reshape_1', 'mrcnn_bbox/Reshape', 'mrcnn_mask/Reshape_1', 'ROI/packed_2', 'rpn_class/concat', 'rpn_bbox/concat']
So all we need is the first four output values i.e: mrcnn_detection, mrcnn_class, mrcnn_bbox, mrcnn_mask.
I am not sure how you got those many values. You can replicate the above print statement and check.
`

[sourcedexter]

I think you are running an older version perhaps. Because. Using the code you have used above, even I am getting 14 tensors for output