The goal of this project is to generate stimuli using generator and generalize the discriminator to predict the variations in translation and stroke of stimuli using random noise.
To Train model from scratch
1. Clone or Download the project
2. Pip install the requirements.txt
3. Set epochs,batch size, model path location
4. Run ModelPipeline.py on python 3
To Predict trained model:
1. Repeat the first three steps as above.
2. Set model path location
2. Run ModelPrediction.py on python 3
Graphical perception technique involves visual decoding of qualitative and quantitative information from the graphs as demonstrated by Cleveland and Mcgill[[1]]. This work is based on Evaluating Graphical perception with CNN where we have experimented the hypothesis (H1.4)[[2]] with GAN. The GAN network [[3]] is treated as deeply convoluted network where generator produces newer representation of visual stimuli and discriminator model would eventually generalize to newer added variations in stimuli.
Our next objective is to make discriminator model generalize to predict the random noise added stimuli. Fine tune model with hyper parameters
[[1]]: http://euclid.psych.yorku.ca/www/psy6135/papers/ClevelandMcGill1984.pdf
[[2]]: https://danielhaehn.com/papers/haehn2018evaluating.pdf
[[3]]: https://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf