Mantis

A lightweight time series classification foundation model used as a frozen representation backbone.

Using Mantis for time series classification

A lightweight time series classification foundation model used as a frozen representation backbone.

In this benchmark, Mantis is evaluated using frozen representation + fitted classifier. Checkpoint information: Mantis-8M. Check the results below for the settings and dataset scopes with published scores.

Evaluated configurations

kNN

MLP

ScaledLR

SVM

Classifier heads are grouped under their shared pretrained model here. They remain separate leaderboard entries because they are different experimental configurations.

Organization: Huawei Noah’s Ark Lab · Université Paris Cité
Parameters: 8.11M backbone
Checkpoint used: Mantis-8M
Input Type: Univariate + multivariate time series
Inference: Frozen representation + fitted classifier

Mantis classification benchmark results

Compare accuracy and average rank only within the same evaluation setting and dataset scope. Standard uses the official training splits; 5-shot and 10-shot use different budget-feasible subsets. Rank and average rank include other models in the full comparison set. Scores are not averaged across configurations.

For comparisons with other backbones, open the foundation model classification leaderboard and use the same setting and scope. The few-shot evaluation protocol explains the label budgets.

Standard · Univariate

Comparison set: 19 configurations

This Standard time series classification comparison reports 5 Mantis configurations for the univariate dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 79.06%, reported by Mantis+SVM with coverage 128 / 128. This configuration ranks 5 of 19 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — Standard Univariate time series classification benchmark results
Configuration	Univariate Rank	Univariate Accuracy (%)	Avg. Rank	Coverage
Mantis+SVM	5	79.06	7.81	128 / 128
Mantis+RF	9	78.67	8.49	128 / 128
Mantis+kNN	12	77.07	9.88	128 / 128
Mantis+NC	15	70.53	12.96	128 / 128
Mantis+MLP	18	63.53	15.76	128 / 128

5-shot · Overall

Comparison set: 18 configurations · Budget-feasible Low-shot subset

This 5-shot time series classification comparison reports 6 Mantis configurations for the overall dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 66.32%, reported by Mantis+ScaledLR with coverage 140 / 140. This configuration ranks 2 of 18 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — 5-shot Overall time series classification benchmark results
Configuration	Overall Rank	Overall Accuracy (%)	Avg. Rank	Coverage
Mantis+ScaledLR	2	66.32	7.15	140 / 140
Mantis+LR	3	66.32	7.24	140 / 140
Mantis+SVM	4	66.22	7.19	140 / 140
Mantis+RF	7	64.02	9.01	140 / 140
Mantis+kNN	8	63.97	9.40	140 / 140
Mantis+NC	13	62.33	10.27	140 / 140

5-shot · Univariate

Comparison set: 19 configurations · Budget-feasible Low-shot subset

This 5-shot time series classification comparison reports 6 Mantis configurations for the univariate dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 67.44%, reported by Mantis+ScaledLR with coverage 112 / 112. This configuration ranks 3 of 19 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — 5-shot Univariate time series classification benchmark results
Configuration	Univariate Rank	Univariate Accuracy (%)	Avg. Rank	Coverage
Mantis+ScaledLR	3	67.44	7.79	112 / 112
Mantis+SVM	4	67.39	7.60	112 / 112
Mantis+LR	5	67.27	7.88	112 / 112
Mantis+RF	8	65.14	9.50	112 / 112
Mantis+kNN	10	65.06	9.94	112 / 112
Mantis+NC	14	62.89	11.22	112 / 112

5-shot · Multivariate

Comparison set: 18 configurations · Budget-feasible Low-shot subset

This 5-shot time series classification comparison reports 6 Mantis configurations for the multivariate dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 62.51%, reported by Mantis+LR with coverage 28 / 28. This configuration ranks 2 of 18 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — 5-shot Multivariate time series classification benchmark results
Configuration	Multivariate Rank	Multivariate Accuracy (%)	Avg. Rank	Coverage
Mantis+LR	2	62.51	6.64	28 / 28
Mantis+ScaledLR	4	61.86	6.59	28 / 28
Mantis+SVM	5	61.54	7.70	28 / 28
Mantis+NC	8	60.08	9.16	28 / 28
Mantis+kNN	9	59.63	9.61	28 / 28
Mantis+RF	10	59.56	9.46	28 / 28

10-shot · Overall

Comparison set: 18 configurations · Budget-feasible Low-shot subset

This 10-shot time series classification comparison reports 6 Mantis configurations for the overall dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 72.07%, reported by Mantis+SVM with coverage 121 / 121. This configuration ranks 1 of 18 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — 10-shot Overall time series classification benchmark results
Configuration	Overall Rank	Overall Accuracy (%)	Avg. Rank	Coverage
Mantis+SVM	1	72.07	6.60	121 / 121
Mantis+ScaledLR	2	71.82	6.71	121 / 121
Mantis+LR	5	71.30	7.32	121 / 121
Mantis+RF	8	69.85	8.86	121 / 121
Mantis+kNN	9	68.71	9.88	121 / 121
Mantis+NC	15	66.26	10.78	121 / 121

10-shot · Univariate

Comparison set: 19 configurations · Budget-feasible Low-shot subset

This 10-shot time series classification comparison reports 6 Mantis configurations for the univariate dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 73.37%, reported by Mantis+ScaledLR with coverage 97 / 97. This configuration ranks 2 of 19 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — 10-shot Univariate time series classification benchmark results
Configuration	Univariate Rank	Univariate Accuracy (%)	Avg. Rank	Coverage
Mantis+ScaledLR	2	73.37	7.13	97 / 97
Mantis+SVM	4	73.22	7.25	97 / 97
Mantis+LR	5	72.72	7.83	97 / 97
Mantis+RF	9	71.05	9.68	97 / 97
Mantis+kNN	12	69.85	10.79	97 / 97
Mantis+NC	16	66.80	11.92	97 / 97

10-shot · Multivariate

Comparison set: 18 configurations · Budget-feasible Low-shot subset

This 10-shot time series classification comparison reports 6 Mantis configurations for the multivariate dataset scope. Each row shows its own coverage.

The highest reported average accuracy among this model's configurations in this setting is 67.42%, reported by Mantis+SVM with coverage 24 / 24. This configuration ranks 2 of 18 in this comparison set. The result describes this configuration and setting, rather than every possible use of the pretrained model.

Mantis — 10-shot Multivariate time series classification benchmark results
Configuration	Multivariate Rank	Multivariate Accuracy (%)	Avg. Rank	Coverage
Mantis+SVM	2	67.42	6.17	24 / 24
Mantis+LR	5	65.57	7.48	24 / 24
Mantis+ScaledLR	6	65.54	7.15	24 / 24
Mantis+RF	7	64.98	8.33	24 / 24
Mantis+kNN	9	64.09	9.00	24 / 24
Mantis+NC	10	64.07	9.25	24 / 24

Mantis paper, code and checkpoint

Reading this model's evaluation

The method catalog describes the pretrained model; leaderboard eligibility depends on the actual coverage of each experimental configuration. Catalog inclusion does not imply complete coverage of all 198 datasets.

Compare evaluated configurations Read the evaluation and ranking rules Explore aggregate capability analysis