Models#

PocketPose provides a collection of pre-trained models for various tasks. You can use them directly for inference in your desktop or mobile applications.

The model zoo is still under construction. More details will be added soon.

Model

Variant

Inputs

Outputs

Precision

Size (MB)

FLOPs (M)

Params (M)

baseline-50

256x192_17

3x192x256

17x48x64

1

32.83

8041.66

23.49

blazepose-full

256x256_33

256x256x3

195, 1, 256x256x1, 64x64x39, 117

2

6.14

774.26

3.17

blazepose-heavy

256x256_33

256x256x3

195, 1, 256x256x1, 64x64x39, 117

2

26.43

3858.81

13.75

blazepose-lite

256x256_33

256x256x3

195, 1, 256x256x1, 64x64x39, 117

2

2.69

397.56

1.36

efficientpose-i

256x256_16

256x256x3

32x32x28, 32x32x16, 32x32x16, 256x256x16

2

1.43

1495.89

0.69

efficientpose-i-lite

256x256_16

256x256x3

256x256x16

2

1.1

1360

0.55

efficientpose-ii

368x368_16

368x368x3

46x46x28, 46x46x16, 46x46x16, 368x368x16

2

3.37

7305.14

1.67

efficientpose-ii-lite

368x368_16

368x368x3

368x368x16

2

2.74

6883.47

1.4

efficientpose-iii

480x480_16

480x480x3

60x60x28, 60x60x16, 60x60x16, 480x480x16

2

6.24

22606.7

3.14

efficientpose-iv

600x600_16

600x600x3

75x75x28, 75x75x16, 75x75x16, 600x600x16

2

12.56

71566.1

6.41

efficientpose-rt

224x224_16

224x224x3

28x28x28, 28x28x16, 28x28x16, 224x224x16

2

0.92

739.09

0.43

efficientpose-rt-lite

224x224_16

224x224x3

224x224x16

2

0.75

721.71

0.37

efficientposenas-a

256x192_17

3x192x256

17x48x64

1

1.49

726.3

1.15

efficientposenas-b

256x192_17

3x192x256

17x48x64

1

3.52

1980.43

3.03

efficientposenas-b

256x256_16

3x256x256

16x64x64

1

3.52

2640.31

3.03

efficientposenas-c

256x192_17

3x192x256

17x48x64

1

5.34

2681.48

4.77

efficientposenas-c

256x256_16

3x256x256

16x64x64

1

5.34

3575.04

4.77

litehrnet-18

256x192_17

3x192x256

17x48x64

1

1.77

406.01

1.11

movenet-lightning

192x192_17

192x192x3

1x17x3

4

8.94

541.61

2.32

movenet-lightning16

192x192_17

192x192x3

1x17x3

2

4.54

541.61

2.32

movenet-lightning8

192x192_17

192x192x3

1x17x3

1

2.76

541.61

2.32

movenet-thunder

256x256_17

256x256x3

1x17x3

4

23.87

2440.7

6.23

movenet-thunder16

256x256_17

256x256x3

1x17x3

2

12

2440.7

6.23

movenet-thunder8

256x256_17

256x256x3

1x17x3

1

6.8

2440.7

6.23

posenet-mobilenet100

257x257_17

257x257x3

9x9x17, 9x9x34, 9x9x32, 9x9x32

4

12.65

1674.53

3.31

res50

256x192_17

3x192x256

17x48x64

1

32.83

8041.66

23.49

Models Timeline

Model

Variant

Inputs

Outputs

Precision

Size (MB)

FLOPs (M)

Params (M)

BlazePose

Full

256x256x3

195, 1, 256x256x1, 64x64x39, 117

2

6.14

774.26

3.17

BlazePose

Heavy

256x256x3

195, 1, 256x256x1, 64x64x39, 117

2

26.43

3858.81

13.75

BlazePose

Lite

256x256x3

195, 1, 256x256x1, 64x64x39, 117

2

2.69

397.56

1.36

EfficientPose

A_COCO

3x192x256

17x48x64

1

1.49

726.3

1.15

EfficientPose

B_COCO

3x192x256

17x48x64

1

3.52

1980.43

3.03

EfficientPose

B_MPII

3x256x256

16x64x64

1

3.52

2640.31

3.03

EfficientPose

C_COCO

3x192x256

17x48x64

1

5.34

2681.48

4.77

EfficientPose

C_MPII

3x256x256

16x64x64

1

5.34

3575.04

4.77

MoveNet

MultiPose-Lightning-FP16

1x1x3

6x56

2

9.14

9.4

4.72

MoveNet

SinglePose-Lightning

192x192x3

1x17x3

4

8.94

541.61

2.32

MoveNet

SinglePose-Lightning-FP16

192x192x3

1x17x3

2

4.54

541.61

2.32

MoveNet

SinglePose-Lightning-INT8

192x192x3

1x17x3

1

2.76

541.61

2.32

MoveNet

SinglePose-Thunder

256x256x3

1x17x3

4

23.87

2440.7

6.23

MoveNet

SinglePose-Thunder-FP16

256x256x3

1x17x3

2

12

2440.7

6.23

MoveNet

SinglePose-Thunder-INT8

256x256x3

1x17x3

1

6.8

2440.7

6.23

PoseNet

MobileNet-075

353x257x3

23x17x17, 23x17x34, 23x17x64, 23x17x1

4

4.82

1358.86

1.26

PoseNet

MobileNet-100

257x257x3

9x9x17, 9x9x34, 9x9x32, 9x9x32

4

12.65

1674.53

3.31

litehrnet

18_coco_256x192

3x192x256

17x48x64

1

1.77

406.01

1.11

litehrnet

18_coco_384x288

3x288x384

17x72x96

1

1.77

913.26

1.11

litehrnet

18_mpii_256x256

3x256x256

16x64x64

1

1.77

540.95

1.11

litehrnet

30_coco_256x192

3x192x256

17x48x64

1

2.83

634.08

1.74

litehrnet

30_coco_384x288

3x288x384

17x72x96

1

2.83

1426.24

1.74

litehrnet

30_mpii_256x256

3x256x256

16x64x64

1

2.83

844.99

1.74

Performance Comparison