Model Zoo

This chapter introduces the Model Zoo of Spacemit.

1 Model List

Classification

Model Name	Fp32 top1/%	Quant top1/%	1-thread inference time @1.6GHz/ms	2-thread inference time @1.6GHz/ms	4-thread inference time @1.6GHz/ms	Model Link
resnet18	69.64	69.64	38.4699	28.289	17.2622
resnet50	75.46	76.00	100.06	71.571	45.1088
seresnet50	79.26	79.00	115.113	90.6925	62.0626
resnext50	80.64	79.28	149.82	126.746	100.67
mobilenet_v1	71.64	71.42	29.9216	21.0166	13.1717
mobilenet_v2	71.32	71.66	34.1042	23.8068	16.0097
mobilenet_v3_large	73.26	73.20	54.3095	35.1863	25.6427
efficientnet_v1_b1	76.7	76.1	114.632	86.3809	56.6777
efficientnet_v2_s	80.3	79.68	149.592	111.917	77.4648
v100_gpu64_6ms	74.58	74.82	57.637	48.2335	35.2046
shufflenet_v2_x1_0	68.72	68.6	37.0776	31.8273	26.3727
inception_v1	66.04	65.1	125.068	85.2864	60.1591
inception_resnet_v2	80.18	80.04	313.24	226.824	142.386
inception_v3	76.74	76.32	121.433	86.5108	55.4845
repvgg	71.82	71.7	24.8253	18.3217	12.5148
squeezenetv1.1	56.12	56.18	18.9487	14.7087	11.3769
swin-tiny_16xb64_in1k	80.26	79.9	1614.54	1273.49	1087.9

Detection

• Detection models include post-processing by default, pre-processing not included

Model Name	Fp32 mAP/%	Quant mAP/%	1-thread inference time @1.6GHz/ms	2-thread inference time @1.6GHz/ms	4-thread inference time @1.6GHz/ms	Model Link
yolov5_n_dyn(320x320)			95.8678	85.5698	67.9318
yolov5_s_dyn(320x320)			150.188	118.762	88.8557
yolov6p5_n(320x320)			40.0114	29.973	23.655
ssd-mobilenet(300x300)			58.1563	43.1459	25.629
nanodet-plus-m(320x320)			84.9497	72.1837	59.2333

Pose

Model Name	Fp32 mAP/%	Quant mAP/%	1-thread inference time @1.6GHz/ms	2-thread inference time @1.6GHz/ms	4-thread inference time @1.6GHz/ms	Model Link
rtmdet-nano			71.4352	54.9239	41.8441
rtmpose-t			36.5269	28.6969	23.0973
rtmpose-s			47.6432	37.6789	28.9216

2 Frequently Asked Questions (FAQ)

Welcome to ask questions