Important Warning: This competition has an experimental format and submission style (images as submission). Competitors must use generative methods to create their submission images and are not permitted to make submissions that include any images already classified as dogs or altered versions of such images.
To enforce and prevent cheating, we reserve the right to: (a) Visually inspect all participants’ submitted images, (b) review any submitted source code, (c) use these reviews to identify violators or determine winners, and (d) disqualify participants from the competition who are found in violation. This is also specified in the competition’s rules
Use your training skills to create images, rather than identify them. You’ll be using GANs, which are at the creative frontier of machine learning. You might think of GANs as robot artists in a sense—able to create eerily lifelike images, and even digital worlds.
“You might not think that programmers are artists, but programming is an extremely creative profession. It’s logic-based creativity. ” – John Romero
A generative adversarial network (GAN) is a class of machine learning system invented by Ian Goodfellow in 2014. Two neural networks compete with each other in a game. Given a training set, this technique learns to generate new data with the same statistics as the training set.
In this competition, you’ll be training generative models to create images of dogs. Only this time… there’s no ground truth data for you to predict. Here, you’ll submit the images and be scored based on how well those images are classified as dogs from pre-trained neural networks. Take these images, for example. Can you tell which are real vs. generated?
rick question; they are all generated!
Why dogs? We chose dogs because, well, who doesn’t love looking at photos of adorable pups? Moreover, dogs can be classified into many sub-categories (breed, color, size), making them ideal candidates for image generation.
Generative methods (in particular, GANs) are currently used in various places on Kaggle for data augmentation. Their potential is vast; they can learn to mimic any distribution of data across any domain: photographs, drawings, music, and prose. If successful, not only will you help advance the state of the art in generative image creation, but you’ll enable us to create more experiments across a variety of domains in the future.
- 1st Place – $2,000
- 2nd Place – $2,000
- 3rd Place – $2,000
- 4th Place – $2,000
- 5th Place – $2,000