MIDI / transcriptionGPU workloadADT_STR · Source adt-str · Weights ADT_STR · tau=0.6 · Paper

Drums to MIDI

ADTOF Plus · ADTOF Direct · ADT_STR · MSG-LD. Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.

15 credits/useFile inputMax 100 MBEstimated 20-1800 seconds

Input

Supports inputs in formats such as mp3 / wav / flac / ogg / m4a / aac / aif / aiff / mp4 / mov / mkv

Supported formats

mp3wavflacoggm4aaacaifaiffmp4movmkv

Output

MIDI

Use case

Edit notes, instrumentation and performance details.

Choose a model family

Drums to MIDI · MSG-LDAvailableSeparate drums and detect onsets to draft drum MIDI; velocity and note playback lengths are fixed.Drums to MIDI · ADTOFAvailableDrums to MIDI: Plus adds velocity and open/closed hi-hat processing; Direct provides the direct transcription path.Drums to MIDI · ADT_STRAvailableTranscribe onsets and velocities for 26 drum classes; the three τ versions are different training setups.

Choose a version or processing mode

Drums to MIDI · ADT_STR · tau=0.8Uses setting-tau-0.8 training weights for 26 drum onsets and velocities. Start with 0.8; compare other versions on the same drum excerpt. τ is not detection sensitivity.RecommendedAvailableDrums to MIDI · ADT_STR · tau=0.4Uses setting-tau-0.4 training weights for 26 drum onsets and velocities. Start with 0.8; compare other versions on the same drum excerpt. τ is not detection sensitivity.AvailableDrums to MIDI · ADT_STR · tau=0.6Uses setting-tau-0.6 training weights for 26 drum onsets and velocities. Start with 0.8; compare other versions on the same drum excerpt. τ is not detection sensitivity.Available

Choose a model for your material and purpose, and review the examples before submitting.

Cost, time, formats and parameters depend on the selected model.

Quick feedback

Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.

Models, parameters, and sources

ADT_STR detects onsets and velocities for 26 drum classes and exports MIDI.

MSG-LD · OBADTOF-PyTorch READMEADTOF (original)ADTOF Direct · PyTorchADTOF Plus paper, Table 1ADTOF PlusOriginal ADTOFADTOF Plus paper, Table 2Vogl et al. 8-class baselineADT-STR · Table 2ADT-STR Setting-1ADT-STR Setting-2

MSG-LD · OB

MSG-LD generates five drum stems and detects onsets; MIDI uses fixed velocity 100 and playback duration 0.1 seconds.

ddman1101/Separate-and-detect MSG-LD · checkpoint MSG-LD

msgld_ob uses the onset branch; msgld_obtb uses both onset and timbre branches.
DDIM sampling steps, 200 by default. Supported values include 50, 100, 200, 250 and 500; some other values are unsupported.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
The paper reports OB, TB and OBTB separately; OB scores do not describe the default OBTB. Note F1 and audio mel-MSE/FAD are separate tables. The paper's detection pipeline has not been remeasured here.

MSG-LD · OBTB

MSG-LD generates five drum stems and detects onsets; MIDI uses fixed velocity 100 and playback duration 0.1 seconds.

ddman1101/Separate-and-detect MSG-LD · checkpoint MSG-LD

msgld_ob uses the onset branch; msgld_obtb uses both onset and timbre branches.
DDIM sampling steps, 200 by default. Supported values include 50, 100, 200, 250 and 500; some other values are unsupported.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
The paper reports OB, TB and OBTB separately; OB scores do not describe the default OBTB. Note F1 and audio mel-MSE/FAD are separate tables. The paper's detection pipeline has not been remeasured here.

ADTOF Plus

ADTOF Plus adds drum separation, velocity mapping and open/closed hi-hat processing to ADTOF transcription and accepts full mixes.

ADTOF Plus source (pinned)ADTOF PyTorch source (pinned)ADTOF Plus paper ADTOF ISMIR 2021 paper Pinned ADTOF Plus technical guide

Enable for a full mix to separate the drums first; disable for an isolated drum track.
Drum onset detection threshold for ADTOF Plus; leave blank to use the model default.
Analyze the original beats after transcription and write the tempo to MIDI, preserving note timing.

ADTOF Direct · PyTorch

ADTOF Direct transcribes isolated drums with PyTorch ADTOF. For full-mix input, TelkNet first extracts drums with AudioSep-Hive; it does not apply Plus velocity or open/closed hi-hat post-processing.

ADTOF PyTorch source (pinned)Original ADTOF source ISMIR 2021 paper materials Signals 2023 paper

Adjust detection thresholds separately for kick, snare, hi-hat, toms and cymbals.
Time resolution, 100 FPS by default; changing it may affect transcription quality.
Analyze the original beats after transcription and write the tempo to MIDI, preserving note timing.

ADT_STR · tau=0.8

pier-maker92/ADT_STR ADT_STR · checkpoint ADT_STR

Select an officially released checkpoint variant.
Maximum generated tokens per segment; the official default is 1024.
Official inference batch size: the number of segments processed together.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Table 2 Setting-1 through Setting-4 are training experiments, not the selectable τ=0.4/0.6/0.8 weights. Scores use isolated drums and do not establish full-mix performance.

ADT_STR · tau=0.4

pier-maker92/ADT_STR ADT_STR · checkpoint ADT_STR

Select an officially released checkpoint variant.
Maximum generated tokens per segment; the official default is 1024.
Official inference batch size: the number of segments processed together.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Table 2 Setting-1 through Setting-4 are training experiments, not the selectable τ=0.4/0.6/0.8 weights. Scores use isolated drums and do not establish full-mix performance.

ADT_STR · tau=0.6

pier-maker92/ADT_STR ADT_STR · checkpoint ADT_STR

Select an officially released checkpoint variant.
Maximum generated tokens per segment; the official default is 1024.
Official inference batch size: the number of segments processed together.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Table 2 Setting-1 through Setting-4 are training experiments, not the selectable τ=0.4/0.6/0.8 weights. Scores use isolated drums and do not establish full-mix performance.

Paper and benchmark notes

Enhanced Automatic Drum Transcription via Drum Stem Source Separation

arXiv - 2025

Paper Pinned implementation

ADTOF: A large dataset of non-synthetic music for automatic drum transcription

ISMIR - 2021

Paper Paper materials

High-Quality and Reproducible Automatic Drum Transcription from Crowdsourced Data

Signals - 2023

Paper

Official / repositories

ddman1101/Separate-and-detect MSG-LD · checkpoint ADTOF Plus source (pinned)ADTOF PyTorch source (pinned)Original ADTOF source pier-maker92/ADT_STR ADT_STR · checkpoint

Papers / technical notes

MSG-LD ADTOF Plus paper ADTOF ISMIR 2021 paper Pinned ADTOF Plus technical guide ISMIR 2021 paper materials Signals 2023 paper ADT_STR

Parameter guide

Enable for a full mix to separate the drums first; disable for an isolated drum track.
Drum onset detection threshold for ADTOF Plus; leave blank to use the model default.
Adjust detection thresholds separately for kick, snare, hi-hat, toms and cymbals.
Time resolution, 100 FPS by default; changing it may affect transcription quality.
Analyze the original beats after transcription and write the tempo to MIDI, preserving note timing.

Workflow

1Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
2Download drum MIDI and review kick, snare, hi-hat and other kit parts in a drum editor.

What to check

Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Time resolution, 100 FPS by default; changing it may affect transcription quality.

Models in this field

3 entries

Includes public test results for models used here and elsewhere. The complete source table is shown by default; conditions and metrics can also be selected for comparison. Scores from different test conditions are not directly comparable.

Only records with verified public inference implementations and matching weights are shown; algorithms requiring no weights are also eligible. Closed services, unreleased weights and unmatched experiment versions are excluded. This is not an all-model leaderboard.

Public weights do not establish a fully open-source license. Expand the data for declared weight licenses and access conditions. Undeclared licenses are marked separately.

BenchmarkRanking category

ADT-STR · ENST / MDB · full vocabulary

Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.

The ADT-STR README recommendation conflicts with its score figure. This ranking uses the values labelled in the official figure.

Rank by metric

ADT-STR · ENST / MDB · full vocabulary
Metric rank	Model	ENST↑F1 (%)	MDB↑F1 (%)
1	ADT-STR · τ=0.4 Public weights · Inference code Public weights Inference code`setting-tau-0.4/model.safetensors`Weight license not declared	60	71
2	ADT-STR · τ=0.6Used on this site Public weights · Inference code Public weights Inference code`setting-tau-0.6/model.safetensors`Weight license not declared	58	65
3	ADT-STR · τ=0.8 Public weights · Inference code Public weights Inference code`setting-tau-0.8/model.safetensors`Weight license not declared	52	61

Original records (3)

Markers identify models or components used here. Scores and settings come from the cited tests, not a new evaluation of the complete site pipeline.

ADT-STR · released checkpoint score figure ↗Checked: 2026-09-27

All evaluation sources (6)

Datasets, versions and conditions remain separate, including sources with no eligible ranking rows.

ADTOF-PyTorch README ↗
Deployable entries 2 / source rows 2
Drums to MIDI ↗
Deployable entries 6 / source rows 6
Drums to MIDI ↗
Deployable entries 3 / source rows 6
Not ranked: 3
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Vogl et al. 8-class baseline · MDB · 8-class evaluation · onset only · 50 ms · Vogl et al. 8-class baseline · RBMA · 8-class evaluation · onset only · 50 ms · Vogl et al. 8-class baseline · ENST · 8-class evaluation · onset only · 50 ms
ADT-STR · Table 2 ↗
Deployable entries 0 / source rows 8
Not ranked: 8
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Setting-1 · ENST · Setting-2 · ENST · Setting-3 · ENST · Setting-4 · ENST · Setting-1 · MDB · Setting-2 · MDB · Setting-3 · MDB · Setting-4 · MDB
MSG-LD · Table 3 ↗
Deployable entries 2 / source rows 3
Not ranked: 1
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
LarsNet
MSG-LD · Table 4 ↗
Deployable entries 2 / source rows 4
Not ranked: 2
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
MSG-LD Vanilla · MSG-LD TB

All original results and sources

Scores apply to the listed dataset, metric and model version; results from different tests are not directly comparable.

The table includes only models with public weights or configurations that can run locally.

ADTOF-PyTorch README

Deployable entries 2 / source rows 2

ADTOF-PyTorch README · PyTorch 88.51 · Keras 88.74

Rank by metric

ADTOF-PyTorch README
Metric rank	Model	Recall↑	Precision↑	F-measure↑
1	ADTOF (original) Public weights · Inference code Frame_RNN_adtofAll_0.data-00000-of-00001 + .index Inference code Weight license (repository declaration): CC-BY-NC-SA-4.0	88.68	89.90	88.74
2	ADTOF-pytorch Public weights · Inference code src/adtof_pytorch/data/adtof_frame_rnn_pytorch_weights.pth Inference codeWeight license not declared	88.26	89.83	88.51

Original records (2)

ADTOF-PyTorch README ↗Checked: 2026-10-25

Drums to MIDI

Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.

Test settingRank by metric

Drums to MIDI
Metric rank	Model	Test setting	F1 (%)↑	Precision (%)↑	Recall (%)↑
1	ADTOF Plus Public weights · Inference code MDX23C 5-stem MDX23C full-mix drums ADTOF PyTorch Inference code`drumsep_5stems_mdx23c_jarredou.ckpt`Weight license (repository declaration): mit	MDB · 5 classes · onset only · 50 ms	89	89	89
2	Original ADTOF Public weights · Inference code Frame_RNN_adtofAll_0.data-00000-of-00001 + .index Inference code Weight license (repository declaration): CC-BY-NC-SA-4.0	MDB · 5 classes · onset only · 50 ms	87	-	-

Original records (6)

ADTOF Plus paper, Table 1 ↗

Drums to MIDI

Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.

Test settingRank by metric

Drums to MIDI
Metric rank	Model	Test setting	F1 (%)↑	Precision (%)↑	Recall (%)↑
-	ADTOF Plus Public weights · Inference code MDX23C 5-stem MDX23C full-mix drums ADTOF PyTorch Inference code`drumsep_5stems_mdx23c_jarredou.ckpt`Weight license (repository declaration): mit	MDB · 8-class evaluation · onset only · 50 ms	84	84	84

Original records (3)

ADTOF Plus paper, Table 2 ↗

MSG-LD · Table 3

Deployable entries 2 / source rows 3

Not ranked: 1

A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.

LarsNet

The paper reports OB, TB and OBTB separately; OB scores do not describe the default OBTB. Note F1 and audio mel-MSE/FAD are separate tables. The paper's detection pipeline has not been remeasured here.

Rank by metric

MSG-LD · Table 3
Metric rank	Model	MDB · Kick			MDB · Snare			MDB · Toms			MDB · Hi-Hats			MDB · Cymbals			MDB · Overall			ENST · Kick			ENST · Snare			ENST · Toms			ENST · Hi-Hats			ENST · Cymbals			ENST · Overall
Metric rank	Model	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑
1	ADTOF Public weights · Inference code Frame_RNN_adtofAll_0.data-00000-of-00001 + .index Inference code Weight license (repository declaration): CC-BY-NC-SA-4.0	0.851	0.790	0.923	0.752	0.828	0.690	0.520	0.520	0.520	0.773	0.859	0.703	0.849	0.802	0.902	0.795	0.813	0.778	0.795	0.984	0.667	0.599	0.932	0.441	0.581	0.669	0.513	0.777	0.936	0.663	0.657	0.788	0.563	0.709	0.908	0.582
2	MSG-LD OB Public weights · Inference code Inference code Weight license (repository declaration): mit	0.931	0.939	0.923	0.760	0.791	0.732	0.268	0.309	0.236	0.650	0.550	0.794	0.481	0.663	0.377	0.707	0.709	0.704	0.821	0.941	0.729	0.642	0.698	0.594	0.491	0.415	0.602	0.630	0.643	0.618	0.391	0.611	0.287	0.640	0.695	0.593

Original records (2)

MSG-LD · Table 3 ↗Checked: 2026-10-25

MSG-LD · Table 4

Deployable entries 2 / source rows 4

Not ranked: 2

A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.

MSG-LD Vanilla · MSG-LD TB

Rank by metric

MSG-LD · Table 4
Metric rank	Model	MDB Overall			ENST Overall			Mean overall F1 (2 sets)
Metric rank	Model	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑
1	MSG-LD OB Public weights · Inference code Inference code Weight license (repository declaration): mit	0.707	0.709	0.704	0.640	0.695	0.593	0.674	0.674	0.674
2	MSG-LD OBTB Public weights · Inference code Inference code Weight license (repository declaration): mit	0.690	0.709	0.672	0.636	0.703	0.580	0.663	0.663	0.663

Original records (2)

MSG-LD · Table 4 ↗Checked: 2026-10-25

MIDI / transcriptionGPU workloadADT_STR · Source adt-str · Weights ADT_STR · tau=0.6 · Paper

Drums to MIDI

ADTOF Plus · ADTOF Direct · ADT_STR · MSG-LD. Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.

15 credits/useFile inputMax 100 MBEstimated 20-1800 seconds

Input

Supports inputs in formats such as mp3 / wav / flac / ogg / m4a / aac / aif / aiff / mp4 / mov / mkv

Supported formats

mp3wavflacoggm4aaacaifaiffmp4movmkv

Output

MIDI

Use case

Edit notes, instrumentation and performance details.

Choose a model family

Choose a version or processing mode

Choose a model for your material and purpose, and review the examples before submitting.

Cost, time, formats and parameters depend on the selected model.

Quick feedback

Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.

Models, parameters, and sources

ADT_STR detects onsets and velocities for 26 drum classes and exports MIDI.

MSG-LD · OB

MSG-LD generates five drum stems and detects onsets; MIDI uses fixed velocity 100 and playback duration 0.1 seconds.

ddman1101/Separate-and-detect MSG-LD · checkpoint MSG-LD

msgld_ob uses the onset branch; msgld_obtb uses both onset and timbre branches.
DDIM sampling steps, 200 by default. Supported values include 50, 100, 200, 250 and 500; some other values are unsupported.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
The paper reports OB, TB and OBTB separately; OB scores do not describe the default OBTB. Note F1 and audio mel-MSE/FAD are separate tables. The paper's detection pipeline has not been remeasured here.

MSG-LD · OBTB

MSG-LD generates five drum stems and detects onsets; MIDI uses fixed velocity 100 and playback duration 0.1 seconds.

ddman1101/Separate-and-detect MSG-LD · checkpoint MSG-LD

msgld_ob uses the onset branch; msgld_obtb uses both onset and timbre branches.
DDIM sampling steps, 200 by default. Supported values include 50, 100, 200, 250 and 500; some other values are unsupported.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
The paper reports OB, TB and OBTB separately; OB scores do not describe the default OBTB. Note F1 and audio mel-MSE/FAD are separate tables. The paper's detection pipeline has not been remeasured here.

ADTOF Plus

ADTOF Plus adds drum separation, velocity mapping and open/closed hi-hat processing to ADTOF transcription and accepts full mixes.

ADTOF Plus source (pinned)ADTOF PyTorch source (pinned)ADTOF Plus paper ADTOF ISMIR 2021 paper Pinned ADTOF Plus technical guide

Enable for a full mix to separate the drums first; disable for an isolated drum track.
Drum onset detection threshold for ADTOF Plus; leave blank to use the model default.
Analyze the original beats after transcription and write the tempo to MIDI, preserving note timing.

ADTOF Direct · PyTorch

ADTOF Direct transcribes isolated drums with PyTorch ADTOF. For full-mix input, TelkNet first extracts drums with AudioSep-Hive; it does not apply Plus velocity or open/closed hi-hat post-processing.

ADTOF PyTorch source (pinned)Original ADTOF source ISMIR 2021 paper materials Signals 2023 paper

Adjust detection thresholds separately for kick, snare, hi-hat, toms and cymbals.
Time resolution, 100 FPS by default; changing it may affect transcription quality.
Analyze the original beats after transcription and write the tempo to MIDI, preserving note timing.

ADT_STR · tau=0.8

pier-maker92/ADT_STR ADT_STR · checkpoint ADT_STR

Select an officially released checkpoint variant.
Maximum generated tokens per segment; the official default is 1024.
Official inference batch size: the number of segments processed together.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Table 2 Setting-1 through Setting-4 are training experiments, not the selectable τ=0.4/0.6/0.8 weights. Scores use isolated drums and do not establish full-mix performance.

ADT_STR · tau=0.4

pier-maker92/ADT_STR ADT_STR · checkpoint ADT_STR

Select an officially released checkpoint variant.
Maximum generated tokens per segment; the official default is 1024.
Official inference batch size: the number of segments processed together.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Table 2 Setting-1 through Setting-4 are training experiments, not the selectable τ=0.4/0.6/0.8 weights. Scores use isolated drums and do not establish full-mix performance.

ADT_STR · tau=0.6

pier-maker92/ADT_STR ADT_STR · checkpoint ADT_STR

Select an officially released checkpoint variant.
Maximum generated tokens per segment; the official default is 1024.
Official inference batch size: the number of segments processed together.
Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Table 2 Setting-1 through Setting-4 are training experiments, not the selectable τ=0.4/0.6/0.8 weights. Scores use isolated drums and do not establish full-mix performance.

Paper and benchmark notes

Enhanced Automatic Drum Transcription via Drum Stem Source Separation

arXiv - 2025

Paper Pinned implementation

ADTOF: A large dataset of non-synthetic music for automatic drum transcription

ISMIR - 2021

Paper Paper materials

High-Quality and Reproducible Automatic Drum Transcription from Crowdsourced Data

Signals - 2023

Paper

Official / repositories

ddman1101/Separate-and-detect MSG-LD · checkpoint ADTOF Plus source (pinned)ADTOF PyTorch source (pinned)Original ADTOF source pier-maker92/ADT_STR ADT_STR · checkpoint

Papers / technical notes

MSG-LD ADTOF Plus paper ADTOF ISMIR 2021 paper Pinned ADTOF Plus technical guide ISMIR 2021 paper materials Signals 2023 paper ADT_STR

Parameter guide

Enable for a full mix to separate the drums first; disable for an isolated drum track.
Drum onset detection threshold for ADTOF Plus; leave blank to use the model default.
Adjust detection thresholds separately for kick, snare, hi-hat, toms and cymbals.
Time resolution, 100 FPS by default; changing it may affect transcription quality.
Analyze the original beats after transcription and write the tempo to MIDI, preserving note timing.

Workflow

1Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
2Download drum MIDI and review kick, snare, hi-hat and other kit parts in a drum editor.

What to check

Full mixes go through the model’s drum-extraction pipeline first. Isolated drums go directly to transcription or drum-stem separation.
Time resolution, 100 FPS by default; changing it may affect transcription quality.

Models in this field

3 entries

Public weights do not establish a fully open-source license. Expand the data for declared weight licenses and access conditions. Undeclared licenses are marked separately.

BenchmarkRanking category

ADT-STR · ENST / MDB · full vocabulary

The ADT-STR README recommendation conflicts with its score figure. This ranking uses the values labelled in the official figure.

Rank by metric

ADT-STR · ENST / MDB · full vocabulary
Metric rank	Model	ENST↑F1 (%)	MDB↑F1 (%)
1	ADT-STR · τ=0.4 Public weights · Inference code Public weights Inference code`setting-tau-0.4/model.safetensors`Weight license not declared	60	71
2	ADT-STR · τ=0.6Used on this site Public weights · Inference code Public weights Inference code`setting-tau-0.6/model.safetensors`Weight license not declared	58	65
3	ADT-STR · τ=0.8 Public weights · Inference code Public weights Inference code`setting-tau-0.8/model.safetensors`Weight license not declared	52	61

Original records (3)

Markers identify models or components used here. Scores and settings come from the cited tests, not a new evaluation of the complete site pipeline.

ADT-STR · released checkpoint score figure ↗Checked: 2026-09-27

All evaluation sources (6)

Datasets, versions and conditions remain separate, including sources with no eligible ranking rows.

ADTOF-PyTorch README ↗
Deployable entries 2 / source rows 2
Drums to MIDI ↗
Deployable entries 6 / source rows 6
Drums to MIDI ↗
Deployable entries 3 / source rows 6
Not ranked: 3
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Vogl et al. 8-class baseline · MDB · 8-class evaluation · onset only · 50 ms · Vogl et al. 8-class baseline · RBMA · 8-class evaluation · onset only · 50 ms · Vogl et al. 8-class baseline · ENST · 8-class evaluation · onset only · 50 ms
ADT-STR · Table 2 ↗
Deployable entries 0 / source rows 8
Not ranked: 8
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Setting-1 · ENST · Setting-2 · ENST · Setting-3 · ENST · Setting-4 · ENST · Setting-1 · MDB · Setting-2 · MDB · Setting-3 · MDB · Setting-4 · MDB
MSG-LD · Table 3 ↗
Deployable entries 2 / source rows 3
Not ranked: 1
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
LarsNet
MSG-LD · Table 4 ↗
Deployable entries 2 / source rows 4
Not ranked: 2
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
MSG-LD Vanilla · MSG-LD TB

All original results and sources

Scores apply to the listed dataset, metric and model version; results from different tests are not directly comparable.

The table includes only models with public weights or configurations that can run locally.

ADTOF-PyTorch README

Deployable entries 2 / source rows 2

ADTOF-PyTorch README · PyTorch 88.51 · Keras 88.74

Rank by metric

ADTOF-PyTorch README
Metric rank	Model	Recall↑	Precision↑	F-measure↑
1	ADTOF (original) Public weights · Inference code Frame_RNN_adtofAll_0.data-00000-of-00001 + .index Inference code Weight license (repository declaration): CC-BY-NC-SA-4.0	88.68	89.90	88.74
2	ADTOF-pytorch Public weights · Inference code src/adtof_pytorch/data/adtof_frame_rnn_pytorch_weights.pth Inference codeWeight license not declared	88.26	89.83	88.51

Original records (2)

ADTOF-PyTorch README ↗Checked: 2026-10-25

Drums to MIDI

Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.

Test settingRank by metric

Drums to MIDI
Metric rank	Model	Test setting	F1 (%)↑	Precision (%)↑	Recall (%)↑
1	ADTOF Plus Public weights · Inference code MDX23C 5-stem MDX23C full-mix drums ADTOF PyTorch Inference code`drumsep_5stems_mdx23c_jarredou.ckpt`Weight license (repository declaration): mit	MDB · 5 classes · onset only · 50 ms	89	89	89
2	Original ADTOF Public weights · Inference code Frame_RNN_adtofAll_0.data-00000-of-00001 + .index Inference code Weight license (repository declaration): CC-BY-NC-SA-4.0	MDB · 5 classes · onset only · 50 ms	87	-	-

Original records (6)

ADTOF Plus paper, Table 1 ↗

Drums to MIDI

Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.

Test settingRank by metric

Drums to MIDI
Metric rank	Model	Test setting	F1 (%)↑	Precision (%)↑	Recall (%)↑
-	ADTOF Plus Public weights · Inference code MDX23C 5-stem MDX23C full-mix drums ADTOF PyTorch Inference code`drumsep_5stems_mdx23c_jarredou.ckpt`Weight license (repository declaration): mit	MDB · 8-class evaluation · onset only · 50 ms	84	84	84

Original records (3)

ADTOF Plus paper, Table 2 ↗

MSG-LD · Table 3

Deployable entries 2 / source rows 3

Not ranked: 1

A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.

LarsNet

Rank by metric

MSG-LD · Table 3
Metric rank	Model	MDB · Kick			MDB · Snare			MDB · Toms			MDB · Hi-Hats			MDB · Cymbals			MDB · Overall			ENST · Kick			ENST · Snare			ENST · Toms			ENST · Hi-Hats			ENST · Cymbals			ENST · Overall
Metric rank	Model	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑
1	ADTOF Public weights · Inference code Frame_RNN_adtofAll_0.data-00000-of-00001 + .index Inference code Weight license (repository declaration): CC-BY-NC-SA-4.0	0.851	0.790	0.923	0.752	0.828	0.690	0.520	0.520	0.520	0.773	0.859	0.703	0.849	0.802	0.902	0.795	0.813	0.778	0.795	0.984	0.667	0.599	0.932	0.441	0.581	0.669	0.513	0.777	0.936	0.663	0.657	0.788	0.563	0.709	0.908	0.582
2	MSG-LD OB Public weights · Inference code Inference code Weight license (repository declaration): mit	0.931	0.939	0.923	0.760	0.791	0.732	0.268	0.309	0.236	0.650	0.550	0.794	0.481	0.663	0.377	0.707	0.709	0.704	0.821	0.941	0.729	0.642	0.698	0.594	0.491	0.415	0.602	0.630	0.643	0.618	0.391	0.611	0.287	0.640	0.695	0.593

Original records (2)

MSG-LD · Table 3 ↗Checked: 2026-10-25

MSG-LD · Table 4

Deployable entries 2 / source rows 4

Not ranked: 2

A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.

MSG-LD Vanilla · MSG-LD TB

Rank by metric

MSG-LD · Table 4
Metric rank	Model	MDB Overall			ENST Overall			Mean overall F1 (2 sets)
Metric rank	Model	F1↑	P↑	R↑	F1↑	P↑	R↑	F1↑	P↑	R↑
1	MSG-LD OB Public weights · Inference code Inference code Weight license (repository declaration): mit	0.707	0.709	0.704	0.640	0.695	0.593	0.674	0.674	0.674
2	MSG-LD OBTB Public weights · Inference code Inference code Weight license (repository declaration): mit	0.690	0.709	0.672	0.636	0.703	0.580	0.663	0.663	0.663

Original records (2)

MSG-LD · Table 4 ↗Checked: 2026-10-25