Centre of Excellence in Machine Learning

Project 1: Car model analysis with artificial neural networks

Objectives

• Explore possible neural architectures to handle the problem of car model identification using grained features.
• Reproduce the results reported in the paper linked here.

Practical applications

• Paying tolls.
• Video surveillance and car verification.
• Smart cities.
• Mobile applications to instantly show car information.
• Predicting popularity of vehicles.

Datasets

• The Comprehensive Cars (CompCars) dataset
• Stanford dataset
• Common Objects in Context (COCO) dataset
• The CIFAR-10 dataset

Tools/Frameworks

• Caffe.
• Keras with TensorFlow backend.
• Pre-trained tensorflow models
• Tensorflow for poets

Expected Results

• Deliver a trained neural network for car model identification
• Outperform accuracy scores reported by the chinese paper.

Checklist

• Request CompCars database.
• Define development machine and database storage.
• Explore the data and gain insights.
• Evaluate and define preprocessing steps.
• Implementation of the neural network (NN) models.
  • Deep convolutional neural network: Inception.
  • Pulsed-coupled NN.
• Presentation of results in any relevant conference (to define...)

References

• Speed/accuracy trade-offs for modern convolutional object detectors
• View Independent Vehicle Make, Model and Color Recognition Using Convolutional Neural Network.
• Using Convolutional Neural Networks to distinguish vehicle pose and vehicle class.
• A Large-Scale Car Dataset for Fine-Grained Categorization and Verification.
• Three-Dimensional Deformable-Model-Based Localization and Recognition of Road Vehicles.

Comments