Skip to content

Commit ed15dc7

Browse files
WanliZhongWanliZhong
committed
add benchmark result
1 parent 3edad00 commit ed15dc7

File tree

1 file changed

+18
-5
lines changed

1 file changed

+18
-5
lines changed

‎benchmark/README.md‎

Lines changed: 18 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -114,6 +114,7 @@ mean median min input size model
114114
9.68 9.21 7.74 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
115115
9.85 10.63 7.74 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
116116
9.63 9.28 7.74 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
117+
1023.71 1024.90 1016.75 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
117118
```
118119
119120
### Raspberry Pi 4B
@@ -171,6 +172,7 @@ mean median min input size model
171172
223.58 219.82 200.44 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
172173
225.60 243.89 200.44 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
173174
220.97 223.16 193.91 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
175+
16176.34 16377.71 15127.76 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
174176
```
175177
176178
### Jetson Nano B01
@@ -229,12 +231,13 @@ mean median min input size model
229231
243.46 238.98 219.06 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
230232
246.87 256.05 219.06 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
231233
243.37 238.90 219.06 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
234+
12659.89 12689.15 12543.48 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
232235
```
233236
234237
GPU (CUDA-FP32):
235238
<!-- config wechat is excluded due to its api does not support setting backend and target -->
236239
```
237-
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat --cfg_overwrite_backend_target 1
240+
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat:raft --cfg_overwrite_backend_target 1
238241
Benchmarking ...
239242
backend=cv.dnn.DNN_BACKEND_CUDA
240243
target=cv.dnn.DNN_TARGET_CUDA
@@ -265,7 +268,7 @@ mean median min input size model
265268
GPU (CUDA-FP16):
266269
<!-- config wechat is excluded due to its api does not support setting backend and target -->
267270
```
268-
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat --cfg_overwrite_backend_target 2
271+
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat:raft --cfg_overwrite_backend_target 2
269272
Benchmarking ...
270273
backend=cv.dnn.DNN_BACKEND_CUDA
271274
target=cv.dnn.DNN_TARGET_CUDA_FP16
@@ -348,6 +351,7 @@ mean median min input size model
348351
182.90 178.97 161.37 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
349352
184.26 194.43 161.37 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
350353
180.65 180.59 155.36 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
354+
9608.99 9618.12 9544.66 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
351355
```
352356
353357
NPU (TIMVX):
@@ -433,14 +437,15 @@ mean median min input size model
433437
277.84 262.99 243.87 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
434438
283.02 280.77 243.87 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
435439
279.21 262.55 243.87 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
440+
22969.43 22973.42 22954.32 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
436441
```
437442
438443
NPU (CANN):
439444
440445
<!-- vittrack is excluded due to HardSwish is not supported by CANN backend yet -->
441446
442447
```
443-
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat:crnn:vittrack --model_exclude pose_estimation_mediapipe_2023mar.onnx --cfg_overwrite_backend_target 4
448+
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat:crnn:vittrack:raft --model_exclude pose_estimation_mediapipe_2023mar.onnx --cfg_overwrite_backend_target 4
444449
Benchmarking ...
445450
backend=cv.dnn.DNN_BACKEND_CANN
446451
target=cv.dnn.DNN_TARGET_NPU
@@ -516,6 +521,7 @@ mean median min input size model
516521
2035.98 2185.05 1268.94 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
517522
1927.93 2178.84 1268.94 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
518523
1822.23 2213.30 1183.93 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
524+
316818.19 316973.10 316458.29 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
519525
```
520526
521527
### Khadas Edge2 (with RK3588)
@@ -574,6 +580,7 @@ mean median min input size model
574580
67.36 65.65 61.13 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
575581
68.52 69.93 61.13 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
576582
68.36 65.65 61.13 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
583+
4643.54 4649.53 4575.53 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
577584
```
578585
579586
### Horizon Sunrise X3 PI
@@ -632,6 +639,7 @@ mean median min input size model
632639
425.24 426.69 380.35 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
633640
431.14 447.85 380.35 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
634641
424.77 417.01 380.35 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
642+
25455.99 25444.32 25274.29 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
635643
```
636644
637645
### MAIX-III AX-PI
@@ -690,6 +698,7 @@ mean median min input size model
690698
3065.33 3217.99 2348.42 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
691699
2976.24 3244.75 2348.42 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
692700
2864.72 3219.46 2208.44 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
701+
433898.12 433937.47 433829.11 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
693702
```
694703
695704
### StarFive VisionFive 2
@@ -747,6 +756,7 @@ mean median min input size model
747756
1313.68 1427.46 808.70 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
748757
1242.07 1408.93 808.70 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
749758
1174.32 1426.07 774.78 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
759+
221318.94 221288.53 221240.32 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
750760
```
751761
752762
### Khadas VIM4
@@ -807,6 +817,7 @@ mean median min input size model
807817
127.63 124.81 113.82 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
808818
129.24 134.50 113.82 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
809819
126.64 125.09 110.45 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
820+
7945.31 7945.83 7917.72 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
810821
```
811822
812823
### Jetson Nano Orin
@@ -865,12 +876,13 @@ mean median min input size model
865876
135.17 130.23 109.24 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
866877
138.38 143.25 109.24 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
867878
137.08 134.22 109.24 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
879+
4404.12 4435.06 4329.78 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
868880
```
869881
870882
GPU (CUDA-FP32):
871883
872884
```
873-
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat --cfg_overwrite_backend_target 1
885+
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat:raft --cfg_overwrite_backend_target 1
874886
Benchmarking ...
875887
backend=cv.dnn.DNN_BACKEND_CUDA
876888
target=cv.dnn.DNN_TARGET_CUDA
@@ -901,7 +913,7 @@ mean median min input size model
901913
GPU (CUDA-FP16):
902914
903915
```
904-
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat --cfg_overwrite_backend_target 2
916+
$ python3 benchmark.py --all --fp32 --cfg_exclude wechat:raft --cfg_overwrite_backend_target 2
905917
Benchmarking ...
906918
backend=cv.dnn.DNN_BACKEND_CUDA
907919
target=cv.dnn.DNN_TARGET_CUDA_FP16
@@ -984,4 +996,5 @@ mean median min input size model
984996
163.43 152.16 135.52 [1280, 720] CRNN with ['text_recognition_CRNN_CH_2022oct_int8.onnx']
985997
173.46 162.85 135.52 [1280, 720] CRNN with ['text_recognition_CRNN_CN_2021nov_int8.onnx']
986998
175.28 145.22 135.52 [1280, 720] CRNN with ['text_recognition_CRNN_EN_2022oct_int8.onnx']
999+
11978.21 11971.35 11135.99 [360, 480] Raft with ['optical_flow_estimation_raft_2023aug.onnx']
9871000
```

0 commit comments

Comments
 (0)