what are the features that are extracted from the last pooling layer of inceptionv3?

Question

what are the features that are extracted from the last pooling layer of inceptionv3? I'm using inceptionv3 to extract features from clothes . I extracted the features from the last pooling layer to use them in another model.

Answer 1

It is composed of 2048 features maps of shape 8x8. I really depends of your input image they don't look all the same. You can try to see them by yourself. For instance, this image (before resizement to 300x300) :

will have features looking like (8 of the 2048 features) :

Here the code used :

model = tf.keras.applications.InceptionV3(include_top=False,weights='imagenet',input_shape=(300,300,3))

img=cv2.resize(cv2.imread('test2.jpg'),(300,300))

pred=model.predict(np.array([img])/255)[0]

fig, axs = plt.subplots(nrows=2, ncols=4, figsize=(15,6))
for i, ax in enumerate(axs.flatten()):
    plt.sca(ax)
    plt.imshow(pred[:,:,i], cmap='gray')
plt.show()