Addition of back-end functionality for Feature-Squeezing

[mr-sarthakgupta]

Addition of back-end functionality for Feature-Squeezing

Task - Create a new file with a function implementing the Feature-Squeezing technique for detecting whether the uploaded image is adversarially perturbed or not.

Difficulty Level - Hard

Steps to be followed -

• Create a new file named "detection.py"
• In the newly created file, add a function named "feature_squeeze" which would take as input an image and would output a boolean value indicating whether the input image is adversarially perturbed. It should be following the framework -

def feature_squeeze():
      model body
      return boolean

• For creating the "feature_squeeze" function refer to the paper Feature Squeezing: Detecting Adversarial Examples in Deep Neural Networks
• Create the app-route python decorator for the algorithm in the app.py file.
• The function should be of the following framework -

@app.route('/cwl2', methods=['GET'])
def cwl2():
    return render_template('cwl2.html')

Resources to be followed -

• Relevant Paper : Feature Squeezing: Detecting Adversarial Examples in Deep Neural Networks