It is a programming project in deep learning, particularly in CNN
ICS 504-Course Project Course Project 1. The goal of the project is to predict the following from images of the faces provided in the dataset: i. Age: is an integer from 0 to 116 ii. Gender: is either 0 (male) or 1 (female) iii. Race: is an integer from 0 to 4, denoting White, Black, Asian, Indian, and Others (like Hispanic, Latino, Middle Eastern), respectively. 2. Download the dataset from: https://www.dropbox.com/s/9nzsrl2n1610hjp/FaceDataset.zip?dl=0 3. Please make sure to randomly assign 15% of the training set as validation data for hyper-parameter tuning and another 15% as test data. 4. Use random seed '777' wherever needed (like taking out 15% data from training set as validation and test set) for reproducibility. 5. Important measures: MAE and MSE for age, Accuracy and F1-score for gender and race. (bold ones are the most important) You need to Experiment with the following setups: 1. [45 points] Try separate model to predict each, i.e., age, gender, and race. You need to perform hyper parameter calibration and fine tuning including different regularization techniques (like dropout, batch-normalization) and to report all the findings/conclusions. 2. [10 points] Apply data augmentation techniques and see the difference. 3. [10 points] Apply transfer learning by using pre-trained models. 4. [10 points] Combining 2 and 3. 5. [20 points] Merging all output variable into one single model with multiple heads to the model. 6. [5 points] Have a mechanism in your code so that if I provide you input test images (new images), it can use them to predict the age, gender and race and show them as output. Extras [10 points]: 1. [5 points] Show class activation maps. 2. [5 points] Other ideas that you may find interesting. + [15 points] The best performing model on the hidden test set will get additional 5% for each type of prediction, i.e., age, gender, and race. Note: ∙ The project should satisfactorily cover the following aspects: error analysis and possible improvements, final results on the test set, and conclusions from the work. ∙ You can work on the project alone or as a maximum of two members. Please send your project member details ASAP by email. ∙ If you are interested in some other topic, please consult your instructor. ∙ If you are interested in a research project, please consult your instructor. ∙ Your work will be checked with appropriate plagiarism detection tools like iThenticate. ∙ You can format your paper as a report or a scientific paper ∙ All the documents (code and report) should be submitted in Jupyter notebooks OR Jupyter notebooks (code) + PDF document (report). ∙ Project presentations will be held in week 15 or 16.