16889_hw2

Late Days Used: 7 days

1.1

1.1

Problem 1.1

1.1

Problem 1.2

1.1

Problem 1.3

1.1

Problem 2.1

Avg F1@0.05: 80.306

Input

1.1

1.1

1.1

Predictions

1.1

1.1

1.1

Ground Truth

1.1

1.1

1.1

Problem 2.2

Avg F1@0.05: 92.754

Input

1.1

1.1

1.1

Predictions

1.1

1.1

1.1

Ground Truth

1.1

1.1

1.1

Problem 2.3

Avg F1@0.05: 86.790

Input

1.1

1.1

1.1

Predictions

1.1

1.1

1.1

Ground Truth

1.1

1.1

1.1

Problem 2.4

Voxel: Avg F1@0.05: 80.306

Pointcloud: Avg F1@0.05: 92.754

Mesh: Avg F1@0.05: 86.790

When we predict pointcloud, we do not need to do any smoothing and interpolation, therefore we have more accurate predictions and higher performance.

Problem 2.5

Larger n_points will lead to higher performance and longer training time.

To be more specific, when I increase the value of n_points from 5000 to 10000 (train for the same number of iterations), the F1-score of mesh prediction increased by 2%.