CPU: ncnn, ONNXRuntime, OpenVINO
GPU: ncnn, TensorRT, PPLNN
- Ubuntu 18.04
- ncnn 20211208
- Cuda 11.3
- TensorRT 7.2.3.4
- Docker 20.10.8
- NVIDIA tesla T4 tensor core GPU for TensorRT.
- Static graph
- Batch size 1
- Synchronize devices after each inference.
- We count the average inference performance of 100 images of the dataset.
- Warm up. For ncnn, we warm up 30 iters for all codebases. As for other backends: for classification, we warm up 1010 iters; for other codebases, we warm up 10 iters.
- Input resolution varies for different datasets of different codebases. All inputs are real images except for
mmediting
because the dataset is not large enough.
Users can directly test the speed through how_to_measure_performance_of_models.md. And here is the benchmark in our environment.
MMCls
MMCls |
TensorRT |
PPLNN |
NCNN |
|
Model |
Dataset |
Input |
fp32 |
fp16 |
int8 |
fp16 |
SnapDragon888-fp32 |
Adreno660-fp32 |
model config file |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
ResNet |
ImageNet |
1x3x224x224 |
2.97 |
336.90 |
1.26 |
791.89 |
1.21 |
829.66 |
1.30 |
768.28 |
33.91 |
29.49 |
25.93 |
38.57 |
$MMCLS_DIR/configs/resnet/resnet50_b32x8_imagenet.py |
ResNeXt |
ImageNet |
1x3x224x224 |
4.31 |
231.93 |
1.42 |
703.42 |
1.37 |
727.42 |
1.36 |
737.67 |
133.44 |
7.49 |
69.38 |
14.41 |
$MMCLS_DIR/configs/resnext/resnext50_32x4d_b32x8_imagenet.py |
SE-ResNet |
ImageNet |
1x3x224x224 |
3.41 |
293.64 |
1.66 |
600.73 |
1.51 |
662.90 |
1.91 |
524.07 |
107.84 |
9.27 |
80.85 |
12.37 |
$MMCLS_DIR/configs/seresnet/seresnet50_b32x8_imagenet.py |
ShuffleNetV2 |
ImageNet |
1x3x224x224 |
1.37 |
727.94 |
1.19 |
841.36 |
1.13 |
883.47 |
4.69 |
213.33 |
9.55 |
104.71 |
10.66 |
93.81 |
$MMCLS_DIR/configs/shufflenet_v2/shufflenet_v2_1x_b64x16_linearlr_bn_nowd_imagenet.py |
MMDet
MMDet |
TensorRT |
PPLNN |
|
Model |
Dataset |
Input |
fp32 |
fp16 |
int8 |
fp16 |
model config file |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
YOLOv3 |
COCO |
1x3x320x320 |
14.76 |
67.76 |
24.92 |
40.13 |
24.92 |
40.13 |
18.07 |
55.35 |
$MMDET_DIR/configs/yolo/yolov3_d53_320_273e_coco.py |
SSD-Lite |
COCO |
1x3x320x320 |
8.84 |
113.12 |
9.21 |
108.56 |
8.04 |
124.38 |
19.72 |
50.71 |
$MMDET_DIR/configs/ssd/ssdlite_mobilenetv2_scratch_600e_coco.py |
RetinaNet |
COCO |
1x3x800x1344 |
97.09 |
10.30 |
25.79 |
38.78 |
16.88 |
59.23 |
38.34 |
26.08 |
$MMDET_DIR/configs/retinanet/retinanet_r50_fpn_1x_coco.py |
FCOS |
COCO |
1x3x800x1344 |
84.06 |
11.90 |
23.15 |
43.20 |
17.68 |
56.57 |
- |
- |
$MMDET_DIR/configs/fcos/fcos_r50_caffe_fpn_gn-head_1x_coco.py |
FSAF |
COCO |
1x3x800x1344 |
82.96 |
12.05 |
21.02 |
47.58 |
13.50 |
74.08 |
30.41 |
32.89 |
$MMDET_DIR/configs/fsaf/fsaf_r50_fpn_1x_coco.py |
Faster-RCNN |
COCO |
1x3x800x1344 |
88.08 |
11.35 |
26.52 |
37.70 |
19.14 |
52.23 |
65.40 |
15.29 |
$MMDET_DIR/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py |
Mask-RCNN |
COCO |
1x3x800x1344 |
320.86 |
3.12 |
241.32 |
4.14 |
- |
- |
86.80 |
11.52 |
$MMDET_DIR/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py |
MMDet |
NCNN |
|
Model |
Dataset |
Input |
SnapDragon888-fp32 |
Adreno660-fp32 |
model config file |
latency (ms) |
FPS |
latency (ms) |
FPS |
MobileNetv2-YOLOv3 |
COCO |
1x3x320x320 |
48.57 |
20.59 |
66.55 |
15.03 |
$MMDET_DIR/configs/yolo/yolov3_mobilenetv2_mstrain-416_300e_coco.py |
SSD-Lite |
COCO |
1x3x320x320 |
44.91 |
22.27 |
66.19 |
15.11 |
$MMDET_DIR/configs/ssd/ssdlite_mobilenetv2_scratch_600e_coco.py |
MMEdit
MMEdit |
TensorRT |
PPLNN |
|
Model |
Input |
fp32 |
fp16 |
int8 |
fp16 |
model config file |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
ESRGAN |
1x3x32x32 |
12.64 |
79.14 |
12.42 |
80.50 |
12.45 |
80.35 |
7.67 |
130.39 |
$MMEDIT_DIR/configs/restorers/esrgan/esrgan_psnr_x4c64b23g32_g1_1000k_div2k.py |
SRCNN |
1x3x32x32 |
0.70 |
1436.47 |
0.35 |
2836.62 |
0.26 |
3850.45 |
0.56 |
1775.11 |
$MMEDIT_DIR/configs/restorers/srcnn/srcnn_x4k915_g1_1000k_div2k.py |
MMOCR
MMOCR |
TensorRT |
PPLNN |
NCNN |
|
Model |
Dataset |
Input |
fp32 |
fp16 |
int8 |
fp16 |
SnapDragon888-fp32 |
Adreno660-fp32 |
model config file |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
DBNet |
ICDAR2015 |
1x3x640x640 |
10.70 |
93.43 |
5.62 |
177.78 |
5.00 |
199.85 |
34.84 |
28.70 |
- |
- |
- |
- |
$MMOCR_DIR/configs/textdet/dbnet/dbnet_r18_fpnc_1200e_icdar2015.py |
CRNN |
IIIT5K |
1x1x32x32 |
1.93 |
518.28 |
1.40 |
713.88 |
1.36 |
736.79 |
- |
- |
10.57 |
94.64 |
20.00 |
50.00 |
$MMOCR_DIR/configs/textrecog/crnn/crnn_academic_dataset.py |
MMSeg
MMSeg |
TensorRT |
PPLNN |
|
Model |
Dataset |
Input |
fp32 |
fp16 |
int8 |
fp16 |
model config file |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
latency (ms) |
FPS |
FCN |
Cityscapes |
1x3x512x1024 |
128.42 |
7.79 |
23.97 |
41.72 |
18.13 |
55.15 |
27.00 |
37.04 |
$MMSEG_DIR/configs/fcn/fcn_r50-d8_512x1024_40k_cityscapes.py |
PSPNet |
Cityscapes |
1x3x512x1024 |
119.77 |
8.35 |
24.10 |
41.49 |
16.33 |
61.23 |
27.26 |
36.69 |
$MMSEG_DIR/configs/pspnet/pspnet_r50-d8_512x1024_80k_cityscapes.py |
DeepLabV3 |
Cityscapes |
1x3x512x1024 |
226.75 |
4.41 |
31.80 |
31.45 |
19.85 |
50.38 |
36.01 |
27.77 |
$MMSEG_DIR/configs/deeplabv3/deeplabv3_r50-d8_512x1024_80k_cityscapes.py |
DeepLabV3+ |
Cityscapes |
1x3x512x1024 |
151.25 |
6.61 |
47.03 |
21.26 |
50.38 |
26.67 |
34.80 |
28.74 |
$MMSEG_DIR/configs/deeplabv3plus/deeplabv3plus_r50-d8_512x1024_80k_cityscapes.py |
Users can directly test the performance through how_to_evaluate_a_model.md. And here is the benchmark in our environment.
MMCls
MMCls |
PyTorch |
ONNX Runtime |
TensorRT |
PPLNN |
|
Model |
Task |
Metrics |
fp32 |
fp32 |
fp32 |
fp16 |
int8 |
fp16 |
model config file |
ResNet-18 |
Classification |
top-1 |
69.90 |
69.88 |
69.88 |
69.86 |
69.86 |
69.86 |
$MMCLS_DIR/configs/resnet/resnet18_b32x8_imagenet.py |
top-5 |
89.43 |
89.34 |
89.34 |
89.33 |
89.38 |
89.34 |
ResNeXt-50 |
Classification |
top-1 |
77.90 |
77.90 |
77.90 |
- |
77.78 |
77.89 |
$MMCLS_DIR/configs/resnext/resnext50_32x4d_b32x8_imagenet.py |
top-5 |
93.66 |
93.66 |
93.66 |
- |
93.64 |
93.65 |
SE-ResNet-50 |
Classification |
top-1 |
77.74 |
77.74 |
77.74 |
77.75 |
77.63 |
77.73 |
$MMCLS_DIR/configs/resnext/resnext50_32x4d_b32x8_imagenet.py |
top-5 |
93.84 |
93.84 |
93.84 |
93.83 |
93.72 |
93.84 |
ShuffleNetV1 1.0x |
Classification |
top-1 |
68.13 |
68.13 |
68.13 |
68.13 |
67.71 |
68.11 |
$MMCLS_DIR/configs/shufflenet_v1/shufflenet_v1_1x_b64x16_linearlr_bn_nowd_imagenet.py |
top-5 |
87.81 |
87.81 |
87.81 |
87.81 |
87.58 |
87.80 |
ShuffleNetV2 1.0x |
Classification |
top-1 |
69.55 |
69.55 |
69.55 |
69.54 |
69.10 |
69.54 |
$MMCLS_DIR/configs/shufflenet_v2/shufflenet_v2_1x_b64x16_linearlr_bn_nowd_imagenet.py |
top-5 |
88.92 |
88.92 |
88.92 |
88.91 |
88.58 |
88.92 |
MobileNet V2 |
Classification |
top-1 |
71.86 |
71.86 |
71.86 |
71.87 |
70.91 |
71.84 |
$MMEDIT_DIR/configs/restorers/real_esrgan/realesrnet_c64b23g32_12x4_lr2e-4_1000k_df2k_ost.py |
top-5 |
90.42 |
90.42 |
90.42 |
90.40 |
89.85 |
90.41 |
MMDet
MMDet |
Pytorch |
ONNXRuntime |
TensorRT |
PPLNN |
OpenVINO |
|
Model |
Task |
Dataset |
Metrics |
fp32 |
fp32 |
fp32 |
fp16 |
int8 |
fp16 |
fp32 |
model config file |
YOLOV3 |
Object Detection |
COCO2017 |
box AP |
33.7 |
- |
33.5 |
33.5 |
33.5 |
- |
- |
$MMDET_DIR/configs/yolo/yolov3_d53_320_273e_coco.py |
SSD |
Object Detection |
COCO2017 |
box AP |
25.5 |
- |
25.5 |
25.5 |
- |
- |
- |
$MMDET_DIR/configs/ssd/ssd300_coco.py |
RetinaNet |
Object Detection |
COCO2017 |
box AP |
36.5 |
- |
36.4 |
36.4 |
36.3 |
36.5 |
- |
$MMDET_DIR/configs/retinanet/retinanet_r50_fpn_1x_coco.py |
FCOS |
Object Detection |
COCO2017 |
box AP |
36.6 |
- |
36.6 |
36.5 |
- |
- |
- |
$MMDET_DIR/configs/fcos/fcos_r50_caffe_fpn_gn-head_1x_coco.py |
FSAF |
Object Detection |
COCO2017 |
box AP |
37.4 |
- |
37.4 |
37.4 |
37.2 |
37.4 |
- |
$MMDET_DIR/configs/fsaf/fsaf_r50_fpn_1x_coco.py |
YOLOX |
Object Detection |
COCO2017 |
box AP |
40.5 |
- |
40.3 |
40.3 |
29.3 |
- |
- |
$MMDET_DIR/configs/yolox/yolox_s_8x8_300e_coco.py |
Faster R-CNN |
Object Detection |
COCO2017 |
box AP |
37.4 |
- |
37.3 |
37.3 |
37.1 |
37.3 |
- |
$MMDET_DIR/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py |
ATSS |
Object Detection |
COCO2017 |
box AP |
39.4 |
- |
39.4 |
39.4 |
- |
- |
- |
$MMDET_DIR/configs/atss/atss_r50_fpn_1x_coco.py |
Cascade R-CNN |
Object Detection |
COCO2017 |
box AP |
40.4 |
- |
40.4 |
40.4 |
- |
40.4 |
- |
$MMDET_DIR/configs/cascade_rcnn/cascade_rcnn_r50_caffe_fpn_1x_coco.py |
Mask R-CNN |
Instance Segmentation |
COCO2017 |
box AP |
38.2 |
- |
38.1 |
38.1 |
- |
38.0 |
- |
$MMDET_DIR/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py |
mask AP |
34.7 |
- |
33.7 |
33.7 |
- |
- |
- |
MMEdit
MMEdit |
Pytorch |
ONNX Runtime |
TensorRT |
PPLNN |
|
Model |
Task |
Dataset |
Metrics |
fp32 |
fp32 |
fp32 |
fp16 |
int8 |
fp16 |
model config file |
SRCNN |
Super Resolution |
Set5 |
PSNR |
28.4316 |
28.4323 |
28.4323 |
28.4286 |
28.1995 |
28.4311 |
$MMEDIT_DIR/configs/restorers/srcnn/srcnn_x4k915_g1_1000k_div2k.py |
SSIM |
0.8099 |
0.8097 |
0.8097 |
0.8096 |
0.7934 |
0.8096 |
ESRGAN |
Super Resolution |
Set5 |
PSNR |
28.2700 |
28.2592 |
28.2592 |
- |
- |
28.2624 |
$MMEDIT_DIR/configs/restorers/esrgan/esrgan_x4c64b23g32_g1_400k_div2k.py |
SSIM |
0.7778 |
0.7764 |
0.7774 |
- |
- |
0.7765 |
ESRGAN-PSNR |
Super Resolution |
Set5 |
PSNR |
30.6428 |
30.6444 |
30.6430 |
- |
- |
27.0426 |
$MMEDIT_DIR/configs/restorers/esrgan/esrgan_psnr_x4c64b23g32_g1_1000k_div2k.py |
|
0.8559 |
0.8558 |
0.8558 |
- |
- |
0.8557 |
SRGAN |
Super Resolution |
Set5 |
PSNR |
27.9499 |
27.9408 |
27.9408 |
- |
- |
27.9388 |
$MMEDIT_DIR/configs/restorers/srresnet_srgan/srgan_x4c64b16_g1_1000k_div2k.pyy |
SSIM |
0.7846 |
0.7839 |
0.7839 |
- |
- |
0.7839 |
SRResNet |
Super Resolution |
Set5 |
PSNR |
30.2252 |
30.2300 |
30.2300 |
- |
- |
30.2294 |
$MMEDIT_DIR/configs/restorers/srresnet_srgan/msrresnet_x4c64b16_g1_1000k_div2k.py |
|
0.8491 |
0.8488 |
0.8488 |
- |
- |
0.8488 |
Real-ESRNet |
Super Resolution |
Set5 |
PSNR |
28.0297 |
27.7016 |
27.7016 |
- |
- |
27.7049 |
$MMEDIT_DIR/configs/restorers/real_esrgan/realesrnet_c64b23g32_12x4_lr2e-4_1000k_df2k_ost.py |
SSIM |
0.8236 |
0.8122 |
0.8122 |
- |
- |
0.8123 |
EDSR |
Super Resolution |
Set5 |
PSNR |
30.2223 |
30.2214 |
30.2214 |
30.2211 |
30.1383 |
- |
$MMEDIT_DIR/configs/restorers/edsr/edsr_x4c64b16_g1_300k_div2k.py |
SSIM |
0.8500 |
0.8497 |
0.8497 |
0.8497 |
0.8469 |
- |
MMOCR
MMOCR |
Pytorch |
ONNXRuntime |
TensorRT |
PPLNN |
OpenVINO |
|
Model |
Task |
Dataset |
Metrics |
fp32 |
fp32 |
fp32 |
fp16 |
int8 |
fp16 |
fp32 |
model config file |
DBNet* |
TextDetection |
ICDAR2015 |
recall |
0.7310 |
0.7304 |
0.7198 |
0.7179 |
0.7111 |
0.7304 |
0.7309 |
$MMOCR_DIR/configs/textdet/dbnet/dbnet_r18_fpnc_1200e_icdar2015.py |
precision |
0.8714 |
0.8718 |
0.8677 |
0.8674 |
0.8688 |
0.8718 |
0.8714 |
hmean |
0.7950 |
0.7949 |
0.7868 |
0.7856 |
0.7821 |
0.7949 |
0.7950 |
CRNN |
TextRecognition |
IIIT5K |
acc |
0.8067 |
0.8067 |
0.8067 |
0.8063 |
0.8067 |
0.8067 |
- |
$MMOCR_DIR/configs/textrecog/crnn/crnn_academic_dataset.py |
SAR |
TextRecognition |
IIIT5K |
acc |
0.9517 |
0.9287 |
- |
- |
- |
- |
- |
$MMOCR_DIR/configs/textrecog/sar/sar_r31_parallel_decoder_academic.py |
MMSeg
MMSeg |
Pytorch |
ONNXRuntime |
TensorRT |
PPLNN |
|
Model |
Dataset |
Metrics |
fp32 |
fp32 |
fp32 |
fp16 |
int8 |
fp16 |
model config file |
FCN |
Cityscapes |
mIoU |
72.25 |
- |
72.36 |
72.35 |
74.19 |
- |
$MMSEG_DIR/configs/fcn/fcn_r50-d8_512x1024_40k_cityscapes.py |
PSPNet |
Cityscapes |
mIoU |
78.55 |
- |
78.26 |
78.24 |
77.97 |
- |
$MMSEG_DIR/configs/pspnet/pspnet_r50-d8_512x1024_80k_cityscapes.py |
deeplabv3 |
Cityscapes |
mIoU |
79.09 |
- |
79.12 |
79.12 |
78.96 |
- |
$MMSEG_DIR/configs/deeplabv3/deeplabv3_r50-d8_512x1024_40k_cityscapes.py |
deeplabv3+ |
Cityscapes |
mIoU |
79.61 |
- |
79.6 |
79.6 |
79.43 |
- |
$MMSEG_DIR/configs/deeplabv3plus/deeplabv3plus_r50-d8_512x1024_40k_cityscapes.py |
Fast-SCNN |
Cityscapes |
mIoU |
70.96 |
- |
70.93 |
70.92 |
66.0 |
- |
$MMSEG_DIR/configs/fastscnn/fast_scnn_lr0.12_8x4_160k_cityscapes.py |
-
As some datasets contain images with various resolutions in codebase like MMDet. The speed benchmark is gained through static configs in MMDeploy, while the performance benchmark is gained through dynamic ones.
-
Some int8 performance benchmarks of TensorRT require Nvidia cards with tensor core, or the performance would drop heavily.
-
DBNet uses the interpolate mode nearest
in the neck of the model, which TensorRT-7 applies a quite different strategy from Pytorch. To make the repository compatible with TensorRT-7, we rewrite the neck to use the interpolate mode bilinear
which improves final detection performance. To get the matched performance with Pytorch, TensorRT-8+ is recommended, which the interpolate methods are all the same as Pytorch.
-
Mask AP of Mask R-CNN drops by 1% for the backend. The main reason is that the predicted masks are directly interpolated to original image in PyTorch, while they are at first interpolated to the preprocessed input image of the model and then to original image in other backends.