Search⌘ K
Home
Courses
Building Advanced Deep Learning and NLP Projects

Project Creation: Part Three

Explore building a Pokemon classifier by applying transfer learning with a pre-trained ResNet50 model. Learn to load images, make predictions, compile the model with Adam optimizer, fine-tune specific layers by freezing others, and train the model to improve accuracy.

We'll cover the following...

Until now, we have built a model using the pre-trained model as a feature Extractor and connected our own classifier on top of that. This is one of the strategies of transfer learning. Before we move to the fine-tuning, let’s try to do some predictions from the model_new we built in the previous lesson.

Make predictions

To make predictions, we would require a sample image to be loaded. Here, we have taken a Pikachu image; and let’s see what output we will get.

Python 3.5
from keras.applications.resnet50 import preprocess_input



image_path = 'pikachu.png'

img = image.load_img(image_path,target_size = (224,224))

x = image.img_to_array(img)

x = np.expand_dims(x,axis=0)

x = preprocess_input(x)



pred = model_new.predict(x)

print(np.argmax(pred))

Explanation:

  • This is the same code we used in the lesson Use a Pre-Trained Model; the only difference is that we cannot use the decode_predictions because we added our own ...