1 of 10

SegNeuro

2 of 10

3 of 10

4 of 10

Our Approach

  • Perform Semantic Segmentation on the 3D Image
    • Treating the Mask as a boolean Matrix and using it as Ground Truth
  • Use the Semantic Mapping to find the number of seeds
    • Taking the output of the previous step and use an a NN to find the number of seeds
  • Cluster the given points using Gaussian Mixture Models
    • Take the output of semantic segmentation and cluster it using the number of seeds
  • Find the seed positions and orientation from the clusters

​

5 of 10

6 of 10

7 of 10

8 of 10

9 of 10

10 of 10

Future Improvements

  • Include different architectures to improve our results on semantic segmentation.
  • Treating the seed counting problem as a classification problem rather than a regression problem.
  • Introduce a reasonable score for seed’s position.
  • Using Mask R-CNN and use an End-to-End model