In this project we aim to implement a Multi-scale Inceptionv3 approach inspired from the paper
Detecting Cancer Metastases on Gigapixel Pathology Images
We use the CAMELYON 16 Multi Giga Pixel slides data. A set of 22 slides were sampled. Each slide has a corresponding mask which annotates the tumor region. We can access around 9 magnification levels.
| Tissue Image | Mask Image |
|---|---|
 |
 |
- Architectures used: Inceptionv3 (Multi-scale) fine-tuned for layers > 150
- Initial Weights: Image Net
- Sliding Window: 80x80
- Center: 50x50
- Loss used: Binary Cross entropy loss
- Magnifications: (2, 3) and (3, 4)