Open leomccormack opened 3 years ago
Working backwards from the metric evaluation functions, it was possible to reverse engineer the structure of estimates and parse the data manually.
Incase it helps anyone:
size(estimates)
ans =
1 4 13 4
% estimates{id_idx, arr_idx, rec_idx, task_idx}
% id_idx: submission ID
% arr_idx: 1:4 ('Robot Head', 'DICIT', 'Hearing Aids', 'Eigenmike')
% rec_idx: 1:13 (for tasks 1+2, and 1:5 for tasks 3+4)
% task_idx: 1:6
% e.g. Eigenmike, task 4:
estimates{1,4,1,4}
ans =
struct with fields:
source: [1×4 struct]
% where:
estimates{1,4,1,4}.source
ans =
1×4 struct array with fields:
azimuth
elevation
timestamps
ID
range
The elements of ".source" then have the dimensions:
azimuth in radians, elevation in radians (although, actually inclination: pi/2-elevation) range in meters
The read_participants_results() function (to obtain "estimates") is referenced in the following code (in the "main_evaluation() function):
Which would allow new systems/data to be evaluated using the same metrics as used in the paper.
However, this function is not included in the repository, and the structure of "estimates" does not appear to be documented anywhere.
Therefore, would it be possible to add a function that is able to parse new data? or have some documentation regarding the structure of "estimates"?