HongChuang/ultralytics-ascend
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 3

Add MNN benchmarks to Raspberry Pi doc (#17910)

Browse source 
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
This commit is contained in:
Lakshantha Dissanayake 2024-12-02 06:49:32 -08:00 • committed by GitHub
parent 5015d1e01b
commit d59a66f7c2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 26 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

47
docs/en/guides/raspberry-pi.md
View file

@ -142,9 +142,10 @@ YOLO11 benchmarks were run by the Ultralytics team on nine different model forma
We have only included benchmarks for YOLO11n and YOLO11s models because other models sizes are too big to run on the Raspberry Pis and does not offer decent performance.
<div style="text-align: center;">
<img width="800" src="https://github.com/ultralytics/docs/releases/download/0/rpi-yolo11-benchmarks.avif" alt="YOLO11 benchmarks on RPi 5">
</div>
<figure style="text-align: center;">
<img width="800" src="https://github.com/ultralytics/assets/releases/download/v0.0.0/rpi-yolo11-benchmarks.avif" alt="YOLO11 benchmarks on RPi 5">
<figcaption style="font-style: italic; color: gray;">Benchmarked with Ultralytics v8.3.39</figcaption>
</figure>
### Detailed Comparison Table
@ -156,29 +157,33 @@ The below table represents the benchmark results for two different models (YOLO1
| Format | Status | Size on disk (MB) | mAP50-95(B) | Inference time (ms/im) |
|---------------|--------|-------------------|-------------|------------------------|
| PyTorch | ✅ | 5.4 | 0.61 | 524.828 |
| TorchScript | ✅ | 10.5 | 0.6082 | 666.874 |
| ONNX | ✅ | 10.2 | 0.6082 | 181.818 |
| OpenVINO | ✅ | 10.4 | 0.6082 | 530.224 |
| TF SavedModel | ✅ | 25.8 | 0.6082 | 405.964 |
| TF GraphDef | ✅ | 10.3 | 0.6082 | 473.558 |
| TF Lite | ✅ | 10.3 | 0.6082 | 324.158 |
| PaddlePaddle | ✅ | 20.4 | 0.6082 | 644.312 |
| NCNN | ✅ | 10.2 | 0.6106 | 93.938 |
| PyTorch | ✅ | 5.4 | 0.6100 | 405.238 |
| TorchScript | ✅ | 10.5 | 0.6082 | 526.628 |
| ONNX | ✅ | 10.2 | 0.6082 | 168.082 |
| OpenVINO | ✅ | 10.4 | 0.6082 | 81.192 |
| TF SavedModel | ✅ | 25.8 | 0.6082 | 377.968 |
| TF GraphDef | ✅ | 10.3 | 0.6082 | 487.244 |
| TF Lite | ✅ | 10.3 | 0.6082 | 317.398 |
| PaddlePaddle | ✅ | 20.4 | 0.6082 | 561.892 |
| MNN | ✅ | 10.1 | 0.6106 | 112.554 |
| NCNN | ✅ | 10.2 | 0.6106 | 88.026 |
=== "YOLO11s"
| Format | Status | Size on disk (MB) | mAP50-95(B) | Inference time (ms/im) |
|---------------|--------|-------------------|-------------|------------------------|
| PyTorch | ✅ | 18.4 | 0.7526 | 1226.426 |
| TorchScript | ✅ | 36.5 | 0.7416 | 1507.95 |
| ONNX | ✅ | 36.3 | 0.7416 | 415.24 |
| OpenVINO | ✅ | 36.4 | 0.7416 | 1167.102 |
| TF SavedModel | ✅ | 91.1 | 0.7416 | 776.14 |
| TF GraphDef | ✅ | 36.4 | 0.7416 | 1014.396 |
| TF Lite | ✅ | 36.4 | 0.7416 | 845.934 |
| PaddlePaddle | ✅ | 72.5 | 0.7416 | 1567.824 |
| NCNN | ✅ | 36.2 | 0.7419 | 197.358 |
| PyTorch | ✅ | 18.4 | 0.7526 | 1011.60 |
| TorchScript | ✅ | 36.5 | 0.7416 | 1268.502 |
| ONNX | ✅ | 36.3 | 0.7416 | 324.17 |
| OpenVINO | ✅ | 36.4 | 0.7416 | 179.324 |
| TF SavedModel | ✅ | 91.1 | 0.7416 | 714.382 |
| TF GraphDef | ✅ | 36.4 | 0.7416 | 1019.83 |
| TF Lite | ✅ | 36.4 | 0.7416 | 849.86 |
| PaddlePaddle | ✅ | 72.5 | 0.7416 | 1276.34 |
| MNN | ✅ | 36.2 | 0.7409 | 273.032 |
| NCNN | ✅ | 36.2 | 0.7419 | 194.858 |
Benchmarked with Ultralytics `v8.3.39`
## Reproduce Our Results