Prepare self trained model for evaluation

[brunovollmer]

Hey everybody,

I'm currently following this tutorial (https://github.com/Microsoft/CNTK/blob/master/Examples/Image/Detection/FastRCNN/BrainScript/CNTK_FastRCNN_Eval.ipynb) to implement a Fast RCNN Evaluator and for the grocery model everything worked fine. Now I want to run the script with my model, which I trained on my own data. The problem is that the structure of the model is different to the pretrained grocery model and as I'm still a beginner with CNTK I don't know who to convert it to the right structure.

The following informations of the structure of the different models are just print(model).

print(grocery-model):

Before preparation: Composite(features: SequenceOver[][Tensor[3,1000,1000]], rois: SequenceOver[][Tensor[100,4]], roiLabels: SequenceOver[][Tensor[100,17]]) -> Tuple[SequenceOver[][Tensor[100,1]], SequenceOver[][Tensor[1,1]], SequenceOver[][Tensor[100,17]]]

After preparation: Composite(features: Sequence[Tensor[3,1000,1000]], rois: Sequence[Tensor[100,4]]) -> Sequence[Tensor[100,17]]

print(my-model):

Before preparation: Composite(data: Tensor[3,850,850], roi_proposals: Tensor[200,4]) -> Tuple[Tensor[200,7], Tensor[200,28]]

Any ideas or suggestions are appreciated.

Update I think I got a part of it this is my code to convert the input:

    # load trained model
    trained_frcnn_model = load_model(modelPath)

    # find the original features and rois input nodes
    features_node = find_by_name(trained_frcnn_model, "data")
    rois_node = find_by_name(trained_frcnn_model, "rois_proposal")

    #  find the output "z" node
    z_node = find_by_name(trained_frcnn_model, 'drop7')

    # define new input nodes for the features (image) and rois
    image_input = input_variable(shape=(3,850,850), name='features')
    roi_input = input_variable(shape=(200,4), name='rois')

    # Clone the desired layers with fixed weights and place holder for the new input nodes
    cloned_nodes = combine([z_node.owner]).clone(
    CloneMethod.freeze,
    {features_node: placeholder(name='features'), rois_node: placeholder(name='rois')})

    # apply the cloned nodes to the input nodes
    self.model = cloned_nodes(image_input, roi_input)

    print("Model loaded successfully!")

But I still don't know about the z output node because I'm pretty sure drop7 is the wrong one but it can't find a z node.