在对医疗数据集进行微调后可视化 ViT 注意力图

Question

I have imported the Vit-b32 model and fine-tuned it to perform classification task on echo images. Now I want to visualize the attention maps so that I can know on which part of the image the model is focusing for doing the classification task. But I am unable to do it and I am getting an error when I am trying to visualize the attention maps after fine-tuning the model. Below is the code:

!pip install --quiet vit-keras
from vit_keras import vit
vit_model = vit.vit_b32(
        image_size = IMAGE_SIZE,
        activation = 'softmax',
        pretrained = True,
        include_top = False,
        pretrained_top = False,
        classes = 3)

When I try yo visualize the attention map without any finetuning then it is working without any error:

from vit_keras import visualize

    x = test_gen.next()
    image = x[0]
    
    attention_map = visualize.attention_map(model = vit_model, image = image)
    
    # Plot results
    fig, (ax1, ax2) = plt.subplots(ncols = 2)
    ax1.axis('off')
    ax2.axis('off')
    ax1.set_title('Original')
    ax2.set_title('Attention Map')
    _ = ax1.imshow(image)
    _ = ax2.imshow(attention_map)

Now in the below code I have added some classification layers to the model and fine-tuned it:

model = tf.keras.Sequential([
        vit_model,
        tf.keras.layers.Flatten(),
        tf.keras.layers.BatchNormalization(),
        tf.keras.layers.Dense(11, activation = tfa.activations.gelu),
        tf.keras.layers.BatchNormalization(),
        tf.keras.layers.Dense(3, 'softmax')
    ],
    name = 'vision_transformer')

model.summary()

Below is the output of the above cell:

> Model: "vision_transformer"
> _________________________________________________________________ Layer (type)                 Output Shape              Param #   
> ================================================================= vit-b32 (Functional)         (None, 768)               87455232  
> _________________________________________________________________ flatten_1 (Flatten)          (None, 768)               0         
> _________________________________________________________________ batch_normalization_2 (Batch (None, 768)               3072      
> _________________________________________________________________ dense_2 (Dense)              (None, 11)                8459      
> _________________________________________________________________ batch_normalization_3 (Batch (None, 11)                44        
> _________________________________________________________________ dense_3 (Dense)              (None, 3)                 36        
> ================================================================= Total params: 87,466,843 Trainable params: 87,465,285 Non-trainable
> params: 1,558
> _________________________________________________________________

Now I have trained the model on my own medical dataset:

learning_rate = 1e-4

optimizer = tfa.optimizers.RectifiedAdam(learning_rate = learning_rate)

model.compile(optimizer = optimizer, 
              loss = tf.keras.losses.CategoricalCrossentropy(label_smoothing = 0.2), 
              metrics = ['accuracy'])

STEP_SIZE_TRAIN = train_gen.n // train_gen.batch_size
STEP_SIZE_VALID = valid_gen.n // valid_gen.batch_size

reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(monitor = 'val_accuracy',
                                                 factor = 0.2,
                                                 patience = 2,
                                                 verbose = 1,
                                                 min_delta = 1e-4,
                                                 min_lr = 1e-6,
                                                 mode = 'max')

earlystopping = tf.keras.callbacks.EarlyStopping(monitor = 'val_accuracy',
                                                 min_delta = 1e-4,
                                                 patience = 5,
                                                 mode = 'max',
                                                 restore_best_weights = True,
                                                 verbose = 1)

checkpointer = tf.keras.callbacks.ModelCheckpoint(filepath = './model.hdf5',
                                                  monitor = 'val_accuracy', 
                                                  verbose = 1, 
                                                  save_best_only = True,
                                                  save_weights_only = True,
                                                  mode = 'max')

callbacks = [earlystopping, reduce_lr, checkpointer]

model.fit(x = train_gen,
          steps_per_epoch = STEP_SIZE_TRAIN,
          validation_data = valid_gen,
          validation_steps = STEP_SIZE_VALID,
          epochs = EPOCHS,
          callbacks = callbacks)

model.save('model.h5', save_weights_only = True)

After training when I am trying to visualize the attention map of the model, it is showing error:

from vit_keras import visualize

x = test_gen.next()
image = x[0]

attention_map = visualize.attention_map(model = model, image = image)

# Plot results
fig, (ax1, ax2) = plt.subplots(ncols = 2)
ax1.axis('off')
ax2.axis('off')
ax1.set_title('Original')
ax2.set_title('Attention Map')
_ = ax1.imshow(image)
_ = ax2.imshow(attention_map)

Below is the following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-13-f208f2d2b771> in <module>
      4 image = x[0]
      5 
----> 6 attention_map = visualize.attention_map(model = model, image = image)
      7 
      8 # Plot results

/opt/conda/lib/python3.7/site-packages/vit_keras/visualize.py in attention_map(model, image)
     14     """
     15     size = model.input_shape[1]
---> 16     grid_size = int(np.sqrt(model.layers[5].output_shape[0][-2] - 1))
     17 
     18     # Prepare the input

TypeError: 'NoneType' object is not subscriptable

Please suggest some way to rectify the above error and visualize the attention maps of the fine-tuned model

Answer 1

You can visualize attention maps by doing the following.

attention_map = visualize.attention_map(model=model.layers[0], image=image)

Since attention_map assumes a ViT model as the model argument, you need to specify the first element of the fine-tuned model defined as tf.keras.Sequential.

Answer 2

maybe it's a little late, but I have a solution working.

I have the path to the image in a string, open it with OpenCv library, and I previosly load a ViT model fine tunned.

I think you only need to use the method get_layer, and select your Vit, as you use it entirely in your sequential model, it works as a layer.

path='/content/drive/MyDrive/TFM/Harvard_procesado/ISIC_0025612.jpg'
img = cv2.imread(path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
res = cv2.resize(img, dsize=(224,224), interpolation=cv2.INTER_CUBIC)

attention_map1 = visualize.attention_map(model = vit_model_t.get_layer('vit_model'), image = res)

fig = plt.figure(figsize=(20,20))
ax = plt.subplot(1, 2, 1)
ax.axis('off')
ax.set_title('Original')
_ = ax.imshow(res)

ax = plt.subplot(1, 2, 2)
ax.axis('off')
ax.set_title('Attention Map')
_ = ax.imshow(attention_map1)

I hope this help

enter image description here