The Torn Page
Seeing what the fire took.
Deep Learning-Based Complete Image Prediction from Partial or Occluded ImagesUndergraduate thesis, defended 2026
27.60 dBPSNR
Seeing what the fire took.
Deep Learning-Based Complete Image Prediction from Partial or Occluded ImagesUndergraduate thesis, defended 2026
27.60 dBPSNR
Supervised by Dr. Chowdhury Mofizur Rahman, BRAC University.
Given an image with part of it missing or occluded, predict the complete image. The framework does it without a GAN, in a single pass, at roughly 32 ms per 256x256 image.
Co-developed with the thesis team, the framework runs in three stages:
| Part | What it involved |
|---|---|
| Mask generation | The masks the model learns to fill. |
| Zero-shot evaluation | Testing on four unseen datasets. |
| Baselines | Retraining five baselines under matched conditions. |
| Ablations | The study showing a 52.7% cut in mapper validation latent loss. |
27.60 dB
PSNR
0.861
SSIM
~32 ms
per 256x256 image
52.7%
cut in mapper validation latent loss
2nd of 6
on zero-shot average PSNR
Real thesis figures and sample input, masked and output images, a link to the paper or report, and possibly an in-browser demo. Until then, the numbers above are the whole account.

PyTorch