2019:Drum Transcription Results

From MIREX Wiki
Revision as of 08:22, 3 November 2019 by Richard Vogl (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Introduction

The drum transcription task was reintroduced 2016 year after it's first edition in 2005. Two out of the three original datasets used in 2005 are available and have been used for evaluation also this year. For those datasets the results from 2005 may be compared to this years results. Additionally to the two datasets from 2005, three new datasets were used in the evaluation. For training the algorithms, the public training set from 2005 plus additional training data taken from the new datasets was provided to the participants. In the context of this task, only the three most common drum instruments—kick/bass drum (KD,BD), snare drum (SD), and hi-hat (HH)—are considered.

As an addition to the three-instrument-class-task, an eight-instrument-class-task, was introduced in 2018. To this end, a new evaluation and training dataset (MIDI) and new annotations for two datasets already used in the three-class-task (MEDLEY, RBMA) were introduced.

For a more detailed discussion of the subtasks an datasets consult the 2019:Drum Transcription task description page.

Submissions

Abstract Contributors
AR1, 3,4,5 PDF Axel Roebel
RV1 PDF Richard Vogl

3 Class Results

The overall results represent the mean values over all datasets.

Overall

Algorithm mean fm sum fm KD mean fm KD sum fm SD mean fm SD sum fm HH mean fm HH sum fm
RV1 0.69 0.74 0.78 0.78 0.69 0.69 0.53 0.53
AR1 0.65 0.71 0.70 0.70 0.63 0.63 0.54 0.54
AR3 0.64 0.70 0.69 0.69 0.61 0.61 0.53 0.53
AR4 0.66 0.71 0.78 0.78 0.60 0.60 0.50 0.50
AR5 0.65 0.71 0.70 0.70 0.62 0.62 0.53 0.53

download these results as csv

2005 baseline: 0.670 (YGO)

The best overall result from 2005 is only provided to put the current results into perspective. Since the overall result form 2005 was calculated on different datasets it is problematic to compare them directly.

IDMT subset

Algorithm mean fm sum fm KD mean fm KD sum fm SD mean fm SD sum fm HH mean fm HH sum fm
RV1 0.66 0.72 0.75 0.75 0.72 0.72 0.53 0.53
AR1 0.60 0.67 0.61 0.61 0.56 0.56 0.56 0.56
AR3 0.60 0.66 0.61 0.61 0.57 0.57 0.55 0.55
AR4 0.59 0.63 0.73 0.73 0.53 0.53 0.45 0.45
AR5 0.60 0.65 0.64 0.64 0.56 0.56 0.55 0.55

download these results as csv

2005 baseline: 0.753 (CD)

KT subset

Algorithm mean fm sum fm KD mean fm KD sum fm SD mean fm SD sum fm HH mean fm HH sum fm
RV1 0.65 0.68 0.80 0.80 0.68 0.68 0.47 0.47
AR1 0.59 0.61 0.66 0.66 0.64 0.64 0.45 0.45
AR3 0.59 0.61 0.65 0.65 0.61 0.61 0.45 0.45
AR4 0.60 0.61 0.75 0.75 0.61 0.61 0.40 0.40
AR5 0.59 0.62 0.67 0.67 0.64 0.64 0.45 0.45

download these results as csv

2005 baseline: 0.617 (YGO)

RBMA subset

Algorithm mean fm sum fm KD mean fm KD sum fm SD mean fm SD sum fm HH mean fm HH sum fm
RV1 0.72 0.74 0.91 0.91 0.62 0.62 0.56 0.56
AR1 0.67 0.72 0.82 0.82 0.54 0.54 0.57 0.57
AR3 0.67 0.71 0.84 0.84 0.52 0.52 0.56 0.56
AR4 0.69 0.72 0.87 0.87 0.54 0.54 0.56 0.56
AR5 0.68 0.72 0.84 0.84 0.55 0.55 0.57 0.57

download these results as csv

MDB subset

Algorithm mean fm sum fm KD mean fm KD sum fm SD mean fm SD sum fm HH mean fm HH sum fm
RV1 0.64 0.65 0.55 0.55 0.60 0.60 0.57 0.57
AR1 0.62 0.64 0.52 0.52 0.63 0.63 0.60 0.60
AR3 0.60 0.61 0.47 0.47 0.61 0.61 0.59 0.59
AR4 0.64 0.63 0.69 0.69 0.59 0.59 0.55 0.55
AR5 0.61 0.62 0.52 0.52 0.62 0.62 0.60 0.60

download these results as csv

MDB-Drums [1]

GEN subset

Algorithm mean fm sum fm KD mean fm KD sum fm SD mean fm SD sum fm HH mean fm HH sum fm
RV1 0.78 0.81 0.91 0.91 0.80 0.80 0.50 0.50
AR1 0.76 0.79 0.86 0.86 0.78 0.78 0.49 0.49
AR3 0.74 0.77 0.87 0.87 0.75 0.75 0.48 0.48
AR4 0.76 0.79 0.86 0.86 0.75 0.75 0.52 0.52
AR5 0.74 0.77 0.86 0.86 0.74 0.74 0.49 0.49

download these results as csv

8 Class Results

The overall results represent the mean values over all datasets.

Overall

Algorithm mean fm sum fm BD mean fm BD sum fm SD mean fm SD sum fm TT mean fm TT sum fm HH mean fm HH sum fm CY mean fm CY sum fm RD mean fm RD sum fm CB mean fm CB sum fm CL mean fm CL sum fm
RV1-8 0.69 0.75 0.82 0.82 0.60 0.60 0.27 0.27 0.61 0.61 0.52 0.52 0.54 0.54 0.51 0.51 0.58 0.58

download these results as csv


RBMA subset

Algorithm mean fm sum fm BD mean fm BD sum fm SD mean fm SD sum fm TT mean fm TT sum fm HH mean fm HH sum fm CY mean fm CY sum fm RD mean fm RD sum fm CB mean fm CB sum fm CL mean fm CL sum fm
RV1-8 0.56 0.60 0.88 0.88 0.34 0.34 0.19 0.19 0.48 0.48 0.36 0.36 0.58 0.58 0.31 0.31 0.43 0.43

download these results as csv

MDB subset

Algorithm mean fm sum fm BD mean fm BD sum fm SD mean fm SD sum fm TT mean fm TT sum fm HH mean fm HH sum fm CY mean fm CY sum fm RD mean fm RD sum fm CB mean fm CB sum fm CL mean fm CL sum fm
RV1-8 0.72 0.70 0.82 0.82 0.73 0.73 0.17 0.17 0.63 0.63 0.51 0.51 0.49 0.49 0.64 0.64 0.36 0.36

download these results as csv

MDB-Drums [2]

MIDI subset

Algorithm mean fm sum fm BD mean fm BD sum fm SD mean fm SD sum fm TT mean fm TT sum fm HH mean fm HH sum fm CY mean fm CY sum fm RD mean fm RD sum fm CB mean fm CB sum fm CL mean fm CL sum fm
RV1-8 0.77 0.85 0.77 0.77 0.72 0.72 0.45 0.45 0.73 0.73 0.67 0.67 0.54 0.54 0.58 0.58 0.94 0.94

download these results as csv