annotate testdata/evaluation/README @ 338:8a2166088b94
livemode
Initialise pitch and shift vectors to 1.0 instead of randomly -- seems to produce at least no worse results, and has the advantage of producing deterministic output
| author |
Chris Cannam |
| date |
Mon, 29 Jun 2015 16:22:32 +0100 |
| parents |
1b7b31aab82a |
| children |
|
| rev |
line source |
|
Chris@138
|
1
|
|
Chris@193
|
2 * run.sh is our basic Silvet evaluation script. It runs Silvet
|
|
Chris@193
|
3 against single- and multiple-instrument files from the TRIOS dataset
|
|
Chris@193
|
4 using the theoretically best settings for each file, and then
|
|
Chris@193
|
5 evaluates the results against the ground truth and MIREX 2012
|
|
Chris@193
|
6 submission.
|
|
Chris@138
|
7
|
|
Chris@193
|
8 * batchProcessingEvaluate.m evaluates a set of .lab output files
|
|
Chris@193
|
9 against MIDI. The MIDI can be found in ../TRIOS-groundtruth, and some
|
|
Chris@193
|
10 example .lab files in ../TRIOS-mirex2012-matlab and ../TRIOS-silvet.
|
|
Chris@138
|
11
|
|
Chris@193
|
12 * evaluate_lab.yeti evaluates one lab file against another, onsets
|
|
Chris@193
|
13 only, to a given timing tolerance.
|
|
Chris@193
|
14
|
