Model Zoo

CASIA-B

Model	`Rank@1.NM`	`Rank@1.BG`	`Rank@1.CL`	`mAP.NM`	`mAP.BG`	`mAP.CL`	Configuration	Input Size	Download checkpoint
GaitSet(AAAI2019)	95.8(95.0)	90.0(87.2)	75.4(70.4)	85.9	81.4	77.0	gaitset.yaml	64x44	-
GaitPart(CVPR2020)	96.1(96.2)	90.7(91.5)	78.7(78.7)	84.8	81.1	75.5	gaitpart.yaml	64x44	-
GLN*(ECCV2020)	96.4(95.6)	93.1(92.0)	81.0(77.2)	89.3	87.0	82.6	gln_phase1.yaml, gln_phase2.yaml	128x88	-
GaitGL(ICCV2021)	97.4(97.4)	94.5(94.5)	83.8(83.6)	86.2	84.4	79.8	gaitgl.yaml	64x44	-
GaitBase(CVPR2023)	97.6	94.0	77.4	-	-	-	gaitbase_da_casiab.yaml	64x44	-

Model	`Rank@1.NM`	`Rank@1.BG`	`Rank@1.CL`	Input modality	Configuration
GaitSet(AAAI2019)	92.3	86.1	73.36	Silhouette	gaitset.yaml
GaitPart(CVPR2020)	93.1	86.0	75.1	Silhouette	gaitpart.yaml
GaitGL(ICCV2021)	94.2	90.0	81.4	Silhouette	gaitgl.yaml
GaitEdge(ECCV2022)	97.9	96.1	86.4	RGB, Silhouette	phase2_gaitedge.yaml
GaitBase(CVPR2023)	96.5	91.5	78.0	Silhouette	-

Model	Input modality	Rank@1.NM	Rank@1.BG	Rank@1.CL	Rank@1.CR	Rank@1.UB	Rank@1.UF	Rank@1.OC	Rank@1.NT	Rank@1.ALL	Rank@5.ALL	Download checkpoint
PointNet	PointCloud	43.6	37.3	25.7	28.8	19.9	30.1	44.3	27.4	31.3	59.8	-
PointNet++	PointCloud	55.9	52.2	41.6	49.6	47.8	45.9	54.2	52.5	50.8	82.4	-
PointTransformer	PointCloud	53.2	48.1	32.1	43.2	39.1	41.8	47.9	47.1	44.4	76.7	-
SimpleView	PointCloud	72.3	68.8	57.2	63.3	49.2	62.5	79.7	66.5	64.8	85.8	-
LidarGait(CVPR2023)	PointCloud	91.8	88.6	74.6	89	67.5	80.9	94.5	90.4	86.8	96.1	-
DeepGaitV2-P3D	Silhouette	87.5	84.1	53.4	81.3	86.2	84.8	88.5	28.8	82.4	92.6	model
SkeletonGait++(AAAI2024)	Silhouette+Skeleton	89.1	87.2	55.3	85.3	87.3	87.6	91.3	47.9	85.6	95.0	model
SwinGait	Silhouette	86.2	81.3	47.4	79.3	81.6	83.1	85.6	28.5	79.7	91.8	-

Model	`Rank@1`	`mAP`	Configuration	Input Size	Inference Time	Model Size	Download checkpoint
GaitSet(AAAI2019)	87.2(87.1)	92.2	gaitset.yaml	64x44	1m26s	6.31M	-
GaitPart(CVPR2020)	88.6(88.7)	93.3	gaitpart.yaml	64x44	8m04s	3.78M	-
GaitGL(ICCV2021)	89.9(89.7)	93.3	gaitgl.yaml	64x44	5m23s	95.62M	-
GaitBase(CVPR2023)	90.8	-	gaitbase_oumvlp.yaml	64x44	-	-	-
DeepGaitV2-3D	92.0	-	-	64x44	-	-	-

Model	`Rank@1`	Configuration	Input Size	Input modality	Inference Time	Model Size
GaitSet(AAAI2019)	48.4	gaitset.yaml	64x44	Silhouette	-	-
GaitPart(CVPR2020)	47.6	gaitpart.yaml	64x44	Silhouette	-	-
GaitGL(ICCV2021)	41.5	gaitgl.yaml	64x44	Silhouette	-	-
GaitGL(BNNeck)(ICCV2021)	51.7	gaitgl.yaml	64x44	Silhouette	-	-
GaitBase(CVPR2023)	60.1	gaitbase_da_grew.yaml	64x44	Silhouette	-	-
DeepGaitV2-3D	79.4	-	64x44	Silhouette	-	-
SwinGait(Transformer-based)	79.3	swingait3D_B1442C2_GREW.yaml	64x44	Silhouette	-	-
SkeletonGait++(AAAI2024)	85.8	skeletongait++_GREW.yaml	64x44	Silhouette+Skeleton	-	-

Model	`Rank@1`	Configuration	Input Size	Input modality	Download checkpoint
GaitSet(AAAI2019)	36.7	-	64x44	Silhouette	-
GaitPart(CVPR2020)	28.2	-	64x44	Silhouette	-
GaitGL(ICCV2021)	29.7	-	64x44	Silhouette	-
SMPLGait(CVPR 2022)	48.3(46.3)	smplgait.yaml	64x44	Silhouette + SMPL	-
GaitBase(CVPR2023)	64.6	gaitbase_da_gait3d.yaml	64x44	Silhouette	-
DeepGaitV2-P3D	74.4	DeepGaitV2_gait3d.yaml	64x44	Silhouette	-
SwinGait(Transformer-based)	75.0	swingait3D_B1442C2_Gait3D.yaml	64x44	Silhouette	-
SkeletonGait++(AAAI2024)	77.6	skeletongait++_Gait3D.yaml	64x44	Silhouette+Skeleton	-

Model	`Rank@1`	`mAP`	Configuration	Input Size	Download checkpoint
GaitSet(AAAI2019)	55.9	46.7	gaitset_gait3d_parsing.yaml	64x44	model
GaitPart(CVPR2020)	43.0	33.9	gaitpart_gait3d_parsing.yaml	64x44	model
GLN(ECCV2020)	45.7	38.6	gln_phase1_gait3d_parsing.yaml, gln_phase2_gait3d_parsing.yaml	64x44	model
GaitGL(ICCV2021)	47.7	36.2	gaitgl_gait3d_parsing.yaml	64x44	model
SMPLGait(CVPR 2022)	60.6	52.3	smplgait_gait3d_parsing.yaml	64x44	model
GaitBase(CVPR2023)	71.2	64.1	gaitbase_gait3d_parsing_btz32x2_fixed.yaml	64x44	model
ParsingGait(ACMMM2023)	76.2	68.2	parsinggait_gait3d_parsing.yaml	64x44	model

Model	`Rank@1.NM`	`Rank@1.BG`	`Rank@1.CL`	Input size	Configuration
GaitSet(AAAI2019)	82.54	75.26	62.53	64x44	gaitset_casiae.yaml
GaitPart(CVPR2020)	82.92	74.36	60.48	64x44	gaitpart_casiae.yaml
GaitBase(CVPR2023)	91.59	86.65	74.73	64x44	gaitbase_casiae.yaml

The results in the parentheses are mentioned in the papers.

Note:

The checkpoint files can be found here.
GaitBase and DeepGaitV2 series are the officially recommended CNN-based baseline models.
SwinGait series are the officially recommended Transformer-based baseline models.
All results exclude identical-view cases in CASIA-B(*) and OUMVLP.
The shown result of GLN is implemented without the compact block.
The mAP is calculated by the new evaluation protocol proposed in A Comprehensive Study on the Evaluation of Silhouette-based Gait Recognition.