Can we use computer vision algorithms to segment out the lips and both the eyes from Celeb dataset.
In this section I have explained my proposed solution and in the further section I have given the extensive analysis of results.
For data preprocessing, I combined the four the images of (left eye, right eye, upper lip and lower lip) into one image. So that we can train our model in end to end fashion.
For Deep Learning Model here I have used Vanila UNet? model for segmentation. In which our input image size was 256 \times 256 \times 3. For calculating the loss I used Dice Loss and for metric calculation I have use IOU( Intersection Over Union).
The hyperparameters which I have used in this assignment are
- Learning Rate: 5e-4
- Optimization: Adam
| Training | Validation | testing | |
|---|---|---|---|
| Dice Loss | 0.084 | 0.098 | 0.098 |
| Mean IOU | 0.776 | 0.763 | 0.762 |
