Utilizing mAP layer in prototxt to output precision metrics

[Ankit Kulshrestha]

As the title states, I have a prototxt file with a layer defined as :

 

layer {
    type: 'Python'
    name: 'mAP'
    bottom: 'bbox-list-scored'
    top: 'mAP'
    top: 'precision'
    top: 'recall'
    python_param {
        module: 'caffe.layers.detectnet.mean_ap'
        layer: 'mAP'
        param_str : '1248, 384, 16'
    }
    include: { phase: TEST  }
}

My questions are:

1. Is there a way I could use the blob of this layer to output the mAP values per image in the input directory?

2. I've tried using the net interface exposed by pycaffe to access the "blob" dictionary. However, net.blobs['mAP'].data[:][:] gives a KeyError. Is mAP layer not defined as a layer in the underlying C++ code of caffe?

3. If I can't use this layer, is there an efficient way of solving question 1? 

Thanks in advance