VQA

Visual Question and Answer on spatial data from a generated synthetic geometrical dataset

The Generator

A dataset is created by running the generator:

python dataset_generator.py

Image files are w

Configuration

The generator is parameterised by the config file config.json

image_dimentions

The image size.

object_size

Range of possible object sizes as a proportion of the image size.

object_count

How many objects in each image. Ramdomly selected from this range.

questions_per_image

The number of questions generated for each image

stop_intersection

If set to true, palcement of shapes is constrained so that bounding boxes don't intersect.

out_dir

The output directory for the dataset.

file_prefix

The image filename prefix.

image_count

The number of images to be generated in the dataset.

thumbnail_scale

Not currently used.