IndexError: index 917 is out of bounds for axis 0 with size 11, on npargmax in tensorflow

Question

I am using the example here to classify images. I have 2147 images belonging to 11 classes. I am running the classifier on the batch of images. So far program is going well but I am not able to assign the labels because I am getting the following error on the code: labels_batch=image_labels[np.argmax(result_batch, axis=-1)]

Traceback (most recent call last): > > labels_batch=image_labels[np.argmax(result_batch, axis=-1)] IndexError: index 917 is out of bounds for axis 0 with size 11

More information: My labels are categories = ["Categorical_Boxplot", "Column_Charts", "Dendogram", "Heatmap", "Line_Chart", "Map","Node-Link_Diagram", "Ordination_Scatterplot", "Pie_Chart", "Scatterplot", "Stacked_Area_Chart"]

and here is how I am trying to assign this label to the classifier

image_labels = np.array(categories)

result_batch = classifier_model.predict(image_batch)
labels_batch = image_labels[np.argmax(result_batch, axis=-1)]
labels_batch

result_batch.shape

(32, 1001)

The shape of my data

Image batch shape: (32, 224, 224, 3) Label batch shape: (32, 11)

I don't know where I am going wrong and how can I fix it? I already tried to append the image_labels to labels_batch instead of assigning it with =. But it did not work.

Classifier:

classifier_url = "https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/classification/2" #@param {type:"string"}
IMAGE_SIZE = hub.get_expected_image_size(hub.Module(classifier_url))
classifier_layer = layers.Lambda(classifier, input_shape=IMAGE_SIZE + [3])
classifier_model = tf.keras.Sequential([classifier_layer])
classifier_model.summary()

that is the problem, if your model has only 11 classes, why would the output has size 1001. Check your model definition. — Danny Fang, May 03 '19 at 08:53
Could you please add your `classifier_model` definition code? I can then help you fix it. — Danny Fang, May 03 '19 at 09:00
I have updated that. I am using the imagenet trained classifier https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/classification/2 and it has 1001 classes. I dont know I can I use it on my dataset with 11 classes — user3050590, May 03 '19 at 09:18

score 1 · Accepted Answer · answered May 03 '19 at 09:23

Your model definition is incomplete. The model from the tutorial has 1001 classes while your model has only 11, so ou should attach a new classfiction head to your model like this.

classifier_url = "https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/classification/2" #@param {type:"string"}
IMAGE_SIZE = hub.get_expected_image_size(hub.Module(classifier_url))
classifier_layer = layers.Lambda(classifier, input_shape=IMAGE_SIZE + [3])
classifier_model = tf.keras.Sequential([classifier_layer, layers.Dense(11, activation='softmax')])
classifier_model.summary()

IndexError: index 917 is out of bounds for axis 0 with size 11, on npargmax in tensorflow

1 Answers1