Currently most object translation tasks within an image require tedious human input. I propose a system that attempts to perform this object translation task from a sentence describing the object and how it is manipulated. I explore and compare different models for this task.
The final report can be found here: autonomous-object-translation.pdf
make_images.py, unit_tests.py, constants.py include the code for generating the dataset, and the train*.ipynb notebooks include the code to create, train, and evaluate the different models (They all use the file my_nn_lib.py which was adapted from code found here). train808GAN.ipynb is still in development.