Project Demo

[francqz31]

Hey nice project , do you have any kind of a video demo to judge the proposed quality?

[iSiddharth20]

Hello @francqz31 ,

I do have a video comparing results from 4 sets of experiments.
Since the available GPU resources and training dataset were limited, I was only able to work with image resolution of 270x480 for 24 Image Sequences (180 Frames Each).

In the video you will see results from 4 sets of experiments:

• Method-1 : Baseline : MSE Loss for AutoEncoder | MSE Loss for LSTM
• Method-2 : Composite Loss ‘A’ for AutoEncoder | MSE Loss for LSTM
• Method-3 : MSE Loss for AutoEncoder | Composite Loss ‘B’ for LSTM
• Method-4 : Proposed Methodology : Composite Loss ‘A’ for AutoEncoder | Composite Loss ‘B’ for LSTM
  • Composite Loss ‘A’ : MSE Loss + 15% MaxEnt Regularization
  • Composite Loss ‘B’ : MSE Loss + 10% SSIM Regularization

It was observed that Proposed Methodology had Better Color Reproduction and Reduced Motion Blur.

Please share your thoughts on the Project and feel free to review Codebase Explanation for more details and Contributing Guidelines if you'd like to contribute to the project.

Feel free to ask any questions you have.

Demo.mp4