Automated Ad Design (A2D2)

Inspiration

The idea of designing a catchy ad takes time, and effort, so our goal was to automate this process to ensure that anyone involved with making or selling products can have an effective presence through engaging ads.

What it does

It analyzes a batch of product images to extract features like _ dominant colors _, and _ wire-frames _ (edge detection) to be fed to the neural network to produce an output image, later on, an input image of a product is given to the model to evaluate a better view of the product (by rotating the object to a proper view to be eye-catching), then an ad design for the input image is generated based on the batch of predefined ads.

How we built it

Using _ image processing _ techniques, along with _ Pytorch _ to create a dense neural network.

Challenges we ran into

Choosing the correct architecture for the neural network, and the algorithm for predicting the output image.

Accomplishments that we're proud of

An input image of a mug with some dirt on was fed to the neural network, then a clean image of a mug in a better angle was generated.

What we learned

Collaboration, and practicing analytical, and creative thinking to solve a problem.

What's next for AdGenerator

Using more sophisticated algorithms to generate output images like _ GANS _ . Also, increasing the number of extracted features from the batch of ads to be later fed to the model.

Built With

• opencv
• python
• pytorch