DL4CV Pre-Training Assessment

Edge detection (the process of finding edges in an image) can be considered as part of any of the following tasks EXCEPT...

• Image segmentation

• Image filtering

• Image enhancement

• Feature extraction

The key aspect towards building a good representation for the set of bird images here is to ...

• find invariances (common features) among the images

• extract the color of the bird correctly

• measure some statistical properties of these images

• eliminate all views that are different from the majority view

Which of the following is NOT a popular Deep Learning architecture?

• ResNet

• Yolo

• RetinaNet

• TMNet

Which of the following is NOT an activation function used in deep learning architectures?

• MinMax

• Sigmoid

• ReLu

• Tanh

What happens during the back-propagation process in neural network optimization?

• The information is propagated from the input layer to the output layer.

• Weights are updated based on the errors propagated from the inferred output.

• The error is calculated and subtracted from the preceding layer.

• The output values are optimized to be closer to the actual answer.

What is the function of the pooling layer in a convolutional neural network (CNN)?

• To group together relevant features in the image into a feature vector.

• To reduce the representation of features into a more compact form.

• To derive finer features that cover a larger scope of area.

• To trim off the excess pixels involved in the convolution process.

To perform transfer learning in deep learning is to...

• Allow the already-trained network to be adapted for other tasks.

• Transfer more data into the deep learning network to continue the training process.

• Pass the gradient of the neurons in the neural network to the next training iteration.

• Transfer the information obtained at the end of the process to be deployed.

Which of the following is not a known strategy to optimize deep learning training more effectively?

• Batch normalization

• Regularization by dropout

• Pre-training step

• Early stopping

What does the region proposal step in standard object detection do?

• Cuts out a sub-area of the image that is likely to contain many objects.

• Find a pool of potential boxes that may contain objects.

• Selects a region that corresponds to an object that is to be detected.

• Automatically selects the background area in the image.

What does semantic segmentation do?

• Label the foreground objects and the background of the image.

• Label each individual and unique object in the image.

• Label the image with multiple categories (labels).

• Label each pixel in the image with a category label.