Error Evaluation

2D Ground-Truth Data:

• Prepare the ground-truth pixel coordinates in a file subject_2D_gt.xlsx.
• Data should include subject name, three targets' pixel coordinates in both cameras.
• Data should be formatted as follows, for example:

Compute Ground-Truth Pose:

• compute_gt_position.py: compute three sets of 3D ground-truth position with three methods:
• Method 1: Use camera 1's data only.
• Method 2: Use camera 2's data only.
• Method 3: Use both cameras' data.
• compute_gt_position_diff.py: compute the euclidean distance between each pair of corresponding 3D ground-truth positions computed by the above three methods.
• compute_gt_normal.py: compute the ground-truth normal vectors based on the 3D ground-truth positions.

Compute Error:

• compute_target_normal.py: compute the predicted target normal vector, based on the predicted target positions.
• compute_position_error.py: compute 3D position error, by the euclidean distance between the predicted position and the ground-truth position.
• compute_normal_error.py: compute 3D orientation error, by the angle difference between the predicted normal vector and the ground-truth normal vector.

K-fold Evaluation:

• k_fold_success_rate.py: Compute success rate mean & std for all thresholds with K-fold validation.
• k_fold_validation.py: K-fold cross validation for the ratio model