Problem 1: Human Image Generation

Approach 1: StyleGAN2 + GFLA

High-Level Architecture for Approach 1
Working of Approach 1

Approach 2: Disentanglement Person Image Generation

High-Level Architecture for Approach 2
Working of Approach 2

Approach 1 v/s Approach 2 Results

Approach 1 v/s Approach 2: Qualitative Results
Approach 1 v/s Approach 2: Quantitative Results

Problem 2: Adding Bags to the models

High-level working of ST-GAN

Experiments with ST-GAN

Experiment 1:

Update through each warp
Some bad results
Some good results

Experiment 2:

Results from Experiment 2

Experiment 3:

Results from Experiment 3

Comparison of all the experiments

Comparison of our experiments with ST-GAN
Loss curves




Step-by-step output 1
Step-by-step output 2
Sample results


Narcissistic Image of a team member showing off that the model works on her picture
The model was able to place the bag correctly on an image of our team member

Future Work



Deep Learning enthusiasts