Heatmap has low resolution (43x73 in the dog example)

[gabrieldemarmiesse]

Hi, thank you for those networks, they're already really useful.

But I have a minor problem. While running the example with the dog, I noticed the heatmap has a very low resolution (43x73). This is the case also with other images (the resultion differ but is still very low) and I tried with the vgg16 and I have the same issue.

But the heatmap works and finds the dog.

Here is the the code i'm using:

if __name__ == "__main__":
    ### Here is a script to compute the heatmap of the dog synsets.
    ## We find the synsets corresponding to dogs on ImageNet website
    s = "n02084071"
    ids = synset_to_dfs_ids(s)
    # Most of the synsets are not in the subset of the synsets used in ImageNet recognition task.
    ids = np.array([id_ for id_ in ids if id_ is not None])

    im = preprocess_image_batch(['examples/dog.jpg'], color_mode="rgb")

    # Test pretrained model
    sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
    model = convnet('alexnet',weights_path="weights/alexnet_weights.h5", heatmap=True)
    model.compile(optimizer=sgd, loss='mse')

    out = model.predict(im)
    heatmap = out[0,ids,:,:].sum(axis=0)

    print(np.shape(heatmap))

    import matplotlib.pyplot as plt
    plt.imsave("heatmap_dog.png",heatmap)

the result:

(43, 73)

Thank you very much for your help.

[gabrieldemarmiesse]

Actually the heatmap of the dog shown as an example has the same resolution. An the heatmap was obtained with a vgg19, that's why I was confused.