Commit ·
49c2c09
0
Parent(s):
First release
Browse files- .gitattributes +43 -0
- README.md +162 -0
- ax_aarch64/ax_yolo11 +3 -0
- axcl_aarch64/axcl_yolo11 +3 -0
- axcl_x86_64/axcl_yolo11 +3 -0
- config.json +0 -0
- cut-onnx.py +8 -0
- football.jpg +3 -0
- ssd_horse.jpg +3 -0
- wm9955/yolo11s.axmodel +3 -0
- wm9955/yolo11x.axmodel +3 -0
- yolo11-test.py +16 -0
- yolo11_axera.py +216 -0
- yolo11_config.json +42 -0
- yolo11_out.jpg +3 -0
- yolo11s-cut.onnx +3 -0
.gitattributes
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
| 2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
| 3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
| 4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
| 5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
| 6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
| 7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
| 8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
| 9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
| 10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
| 11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
| 12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
| 13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
| 14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
| 15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
| 16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
| 17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
| 18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
| 19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
| 20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
| 21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
| 22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
| 23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
| 24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
| 25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
| 26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
| 27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
| 28 |
+
*.tar filter=lfs diff=lfs merge=lfs -text
|
| 29 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
| 30 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
| 31 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
| 32 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
| 33 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
*.axmodel filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
ax_yolo11 filter=lfs diff=lfs merge=lfs -text
|
| 38 |
+
football.jpg filter=lfs diff=lfs merge=lfs -text
|
| 39 |
+
yolo11_out.jpg filter=lfs diff=lfs merge=lfs -text
|
| 40 |
+
ax_aarch64/ax_yolo11 filter=lfs diff=lfs merge=lfs -text
|
| 41 |
+
axcl_aarch64/axcl_yolo11 filter=lfs diff=lfs merge=lfs -text
|
| 42 |
+
axcl_x86_64/axcl_yolo11 filter=lfs diff=lfs merge=lfs -text
|
| 43 |
+
ssd_horse.jpg filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,162 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
base_model:
|
| 6 |
+
- Ultralytics/YOLO11
|
| 7 |
+
pipeline_tag: object-detection
|
| 8 |
+
tags:
|
| 9 |
+
- Ultralytics
|
| 10 |
+
- YOLO11
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# YOLO11
|
| 14 |
+
|
| 15 |
+
This version of YOLO11 has been converted to run on the WaveMatrix NPU using **w8a16** quantization.
|
| 16 |
+
|
| 17 |
+
This model has been optimized with the following LoRA:
|
| 18 |
+
|
| 19 |
+
Compatible with Pulsar2 version: 3.4
|
| 20 |
+
|
| 21 |
+
## Convert tools links:
|
| 22 |
+
|
| 23 |
+
For those who are interested in model conversion, you can try to export axmodel through
|
| 24 |
+
|
| 25 |
+
- [The repo of ax-samples](https://github.com/WaveMatrix/ax-samples), which you can get the how to build the `ax_yolo11`
|
| 26 |
+
|
| 27 |
+
- [The repo of axcl-samples](https://github.com/WaveMatrix/axcl-samples), which you can get the how to build the `axcl_yolo11`
|
| 28 |
+
|
| 29 |
+
- [Pulsar2 Link, How to Convert ONNX to axmodel](https://pulsar2-docs.readthedocs.io/en/latest/pulsar2/introduction.html)
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
## Support Platform
|
| 33 |
+
|
| 34 |
+
- WM9955
|
| 35 |
+
- Demo board
|
| 36 |
+
- M.2 Accelerator card
|
| 37 |
+
|
| 38 |
+
|Chips|cost|
|
| 39 |
+
|--|--|
|
| 40 |
+
|WM9955| 25 ms |
|
| 41 |
+
|
| 42 |
+
## How to use
|
| 43 |
+
|
| 44 |
+
Download all files from this repository to the device
|
| 45 |
+
|
| 46 |
+
```
|
| 47 |
+
(axcl) wavematrix@raspberrypi:~/samples/WaveMatrix/YOLO11 $ tree -L 2
|
| 48 |
+
.
|
| 49 |
+
├── wm9955
|
| 50 |
+
│ ├── yolo11s.axmodel
|
| 51 |
+
│ └── yolo11x.axmodel
|
| 52 |
+
├── ax_aarch64
|
| 53 |
+
│ └── ax_yolo11
|
| 54 |
+
├── axcl_aarch64
|
| 55 |
+
│ └── axcl_yolo11
|
| 56 |
+
├── axcl_x86_64
|
| 57 |
+
│ └── axcl_yolo11
|
| 58 |
+
├── config.json
|
| 59 |
+
├── cut-onnx.py
|
| 60 |
+
├── football.jpg
|
| 61 |
+
├── README.md
|
| 62 |
+
├── ssd_horse.jpg
|
| 63 |
+
├── yolo11_config.json
|
| 64 |
+
├── yolo11_out.jpg
|
| 65 |
+
├── yolo11s-cut.onnx
|
| 66 |
+
└── yolo11-test.py
|
| 67 |
+
|
| 68 |
+
6 directories, 15 files
|
| 69 |
+
|
| 70 |
+
```
|
| 71 |
+
|
| 72 |
+
### Inference
|
| 73 |
+
|
| 74 |
+
Input image:
|
| 75 |
+

|
| 76 |
+
|
| 77 |
+
#### Inference with WM9955 Host, such as demo board
|
| 78 |
+
|
| 79 |
+
```
|
| 80 |
+
root@wm9955:~/samples/WaveMatrix/YOLO11# ./ax_aarch64/ax_yolo11 -m wm9955/yolo11x.axmodel -i football.jpg
|
| 81 |
+
--------------------------------------
|
| 82 |
+
model file : wm9955/yolo11x.axmodel
|
| 83 |
+
image file : football.jpg
|
| 84 |
+
img_h, img_w : 640 640
|
| 85 |
+
--------------------------------------
|
| 86 |
+
Engine creating handle is done.
|
| 87 |
+
Engine creating context is done.
|
| 88 |
+
Engine get io info is done.
|
| 89 |
+
Engine alloc io is done.
|
| 90 |
+
Engine push input is done.
|
| 91 |
+
--------------------------------------
|
| 92 |
+
post process cost time:4.20 ms
|
| 93 |
+
--------------------------------------
|
| 94 |
+
Repeat 1 times, avg time 24.56 ms, max_time 24.56 ms, min_time 24.56 ms
|
| 95 |
+
--------------------------------------
|
| 96 |
+
detection num: 9
|
| 97 |
+
0: 94%, [ 757, 220, 1127, 1154], person
|
| 98 |
+
0: 94%, [ 0, 357, 314, 1112], person
|
| 99 |
+
0: 93%, [1353, 339, 1629, 1037], person
|
| 100 |
+
0: 91%, [ 494, 476, 659, 1001], person
|
| 101 |
+
32: 86%, [1231, 877, 1281, 922], sports ball
|
| 102 |
+
32: 73%, [ 774, 887, 828, 938], sports ball
|
| 103 |
+
32: 66%, [1012, 882, 1051, 927], sports ball
|
| 104 |
+
0: 54%, [ 0, 543, 83, 1000], person
|
| 105 |
+
0: 46%, [1837, 696, 1877, 814], person
|
| 106 |
+
--------------------------------------
|
| 107 |
+
```
|
| 108 |
+
|
| 109 |
+
Output image:
|
| 110 |
+

|
| 111 |
+
|
| 112 |
+
|
| 113 |
+
#### Inference with M.2 Accelerator card
|
| 114 |
+
|
| 115 |
+
```
|
| 116 |
+
(axcl) wavematrix@raspberrypi:~/samples/WaveMatrix/YOLO11 $ ./axcl_aarch64/axcl_yolo11 -m wm9955/yolo11x.axmodel -i football.jpg
|
| 117 |
+
--------------------------------------
|
| 118 |
+
model file : wm9955/yolo11x.axmodel
|
| 119 |
+
image file : football.jpg
|
| 120 |
+
img_h, img_w : 640 640
|
| 121 |
+
--------------------------------------
|
| 122 |
+
axclrtEngineCreateContextt is done.
|
| 123 |
+
axclrtEngineGetIOInfo is done.
|
| 124 |
+
|
| 125 |
+
grpid: 0
|
| 126 |
+
|
| 127 |
+
input size: 1
|
| 128 |
+
name: images
|
| 129 |
+
1 x 640 x 640 x 3
|
| 130 |
+
|
| 131 |
+
|
| 132 |
+
output size: 3
|
| 133 |
+
name: /model.23/Concat_output_0
|
| 134 |
+
1 x 80 x 80 x 144
|
| 135 |
+
|
| 136 |
+
name: /model.23/Concat_1_output_0
|
| 137 |
+
1 x 40 x 40 x 144
|
| 138 |
+
|
| 139 |
+
name: /model.23/Concat_2_output_0
|
| 140 |
+
1 x 20 x 20 x 144
|
| 141 |
+
|
| 142 |
+
==================================================
|
| 143 |
+
|
| 144 |
+
Engine push input is done.
|
| 145 |
+
--------------------------------------
|
| 146 |
+
post process cost time:1.38 ms
|
| 147 |
+
--------------------------------------
|
| 148 |
+
Repeat 1 times, avg time 24.73 ms, max_time 24.73 ms, min_time 24.73 ms
|
| 149 |
+
--------------------------------------
|
| 150 |
+
detection num: 9
|
| 151 |
+
0: 94%, [ 757, 220, 1127, 1154], person
|
| 152 |
+
0: 94%, [ 0, 357, 314, 1112], person
|
| 153 |
+
0: 93%, [1353, 339, 1629, 1037], person
|
| 154 |
+
0: 91%, [ 494, 476, 659, 1001], person
|
| 155 |
+
32: 86%, [1231, 877, 1281, 922], sports ball
|
| 156 |
+
32: 73%, [ 774, 887, 828, 938], sports ball
|
| 157 |
+
32: 66%, [1012, 882, 1051, 927], sports ball
|
| 158 |
+
0: 54%, [ 0, 543, 83, 1000], person
|
| 159 |
+
0: 46%, [1837, 696, 1877, 814], person
|
| 160 |
+
--------------------------------------
|
| 161 |
+
|
| 162 |
+
```
|
ax_aarch64/ax_yolo11
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2498cb336fc589eb89637b0a55b0b075d7a07632d5ae04977efdb5c44389129d
|
| 3 |
+
size 5466632
|
axcl_aarch64/axcl_yolo11
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4dd953dc7f1cd0def3332fc01ac891673d2c4ca4a36ff3d756c7dc8ab7e9572c
|
| 3 |
+
size 135168
|
axcl_x86_64/axcl_yolo11
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b94a71f04fd79c901a6b1ed75552be3d62ee162e9c3e2003656da634129a7479
|
| 3 |
+
size 158664
|
config.json
ADDED
|
File without changes
|
cut-onnx.py
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import onnx
|
| 2 |
+
|
| 3 |
+
input_path = "yolo11s.onnx"
|
| 4 |
+
output_path = "yolo11s-cut.onnx"
|
| 5 |
+
input_names = ["images"]
|
| 6 |
+
output_names = ["/model.23/Concat_output_0", "/model.23/Concat_1_output_0", "/model.23/Concat_2_output_0"]
|
| 7 |
+
|
| 8 |
+
onnx.utils.extract_model(input_path, output_path, input_names, output_names)
|
football.jpg
ADDED
|
Git LFS Details
|
ssd_horse.jpg
ADDED
|
Git LFS Details
|
wm9955/yolo11s.axmodel
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bad83368e9d68ae9740e9bb72c34e6b8173635faca7eb366e5484f578fa10fa2
|
| 3 |
+
size 10491152
|
wm9955/yolo11x.axmodel
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:18e3cd31f8a04750480f5e9c2d98a2684a51c6b5596da2def971427841e65d8b
|
| 3 |
+
size 60111496
|
yolo11-test.py
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from ultralytics import YOLO
|
| 2 |
+
|
| 3 |
+
# Load a model
|
| 4 |
+
model = YOLO("yolo11s.pt")
|
| 5 |
+
model.info()
|
| 6 |
+
|
| 7 |
+
# Use the model
|
| 8 |
+
results = model("./images/ssd_horse.jpg")
|
| 9 |
+
|
| 10 |
+
# Save the results
|
| 11 |
+
results[0].save("yolo11s-result.jpg")
|
| 12 |
+
|
| 13 |
+
# Export to onnx with simplify
|
| 14 |
+
model.export(format='onnx', simplify=True)
|
| 15 |
+
|
| 16 |
+
|
yolo11_axera.py
ADDED
|
@@ -0,0 +1,216 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import axengine as axe
|
| 2 |
+
import numpy as np
|
| 3 |
+
import cv2
|
| 4 |
+
import argparse
|
| 5 |
+
from dataclasses import dataclass
|
| 6 |
+
|
| 7 |
+
# COCO Class Names
|
| 8 |
+
COCO_CLASSES = [
|
| 9 |
+
'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
|
| 10 |
+
'train', 'truck', 'boat', 'traffic light', 'fire hydrant',
|
| 11 |
+
'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog',
|
| 12 |
+
'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra',
|
| 13 |
+
'giraffe', 'backpack', 'umbrella', 'handbag', 'tie',
|
| 14 |
+
'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
|
| 15 |
+
'kite', 'baseball bat', 'baseball glove', 'skateboard',
|
| 16 |
+
'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup',
|
| 17 |
+
'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
|
| 18 |
+
'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza',
|
| 19 |
+
'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed',
|
| 20 |
+
'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
|
| 21 |
+
'keyboard', 'cell phone', 'microwave', 'oven', 'toaster',
|
| 22 |
+
'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors',
|
| 23 |
+
'teddy bear', 'hair drier', 'toothbrush'
|
| 24 |
+
]
|
| 25 |
+
|
| 26 |
+
@dataclass
|
| 27 |
+
class Object:
|
| 28 |
+
bbox: list # [x0, y0, width, height]
|
| 29 |
+
label: int
|
| 30 |
+
prob: float
|
| 31 |
+
|
| 32 |
+
def sigmoid(x):
|
| 33 |
+
return 1 / (1 + np.exp(-x))
|
| 34 |
+
|
| 35 |
+
def softmax(x, axis=-1):
|
| 36 |
+
x = x - np.max(x, axis=axis, keepdims=True)
|
| 37 |
+
e_x = np.exp(x)
|
| 38 |
+
return e_x / np.sum(e_x, axis=axis, keepdims=True)
|
| 39 |
+
|
| 40 |
+
def decode_distributions(feat, reg_max=16):
|
| 41 |
+
prob = softmax(feat, axis=-1)
|
| 42 |
+
dis = np.sum(prob * np.arange(reg_max), axis=-1)
|
| 43 |
+
return dis
|
| 44 |
+
|
| 45 |
+
def preprocess(image_path, input_size):
|
| 46 |
+
image = cv2.imread(image_path)
|
| 47 |
+
if image is None:
|
| 48 |
+
raise FileNotFoundError(f"Unable to read image file: {image_path}")
|
| 49 |
+
original_shape = image.shape[:2]
|
| 50 |
+
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
| 51 |
+
resized_image = cv2.resize(image, input_size)
|
| 52 |
+
input_tensor = np.expand_dims(resized_image, axis=0).astype(np.uint8)
|
| 53 |
+
return input_tensor, original_shape, image
|
| 54 |
+
|
| 55 |
+
def postprocess(outputs, original_shape, input_size, confidence_threshold, nms_threshold, reg_max=16):
|
| 56 |
+
heads = [
|
| 57 |
+
{'output': outputs[0], 'grid_size': input_size[0] // 8, 'stride': 8},
|
| 58 |
+
{'output': outputs[1], 'grid_size': input_size[0] // 16, 'stride': 16},
|
| 59 |
+
{'output': outputs[2], 'grid_size': input_size[0] // 32, 'stride': 32}
|
| 60 |
+
]
|
| 61 |
+
detections = []
|
| 62 |
+
num_classes = 80
|
| 63 |
+
bbox_channels = 4 * reg_max
|
| 64 |
+
class_channels = num_classes
|
| 65 |
+
|
| 66 |
+
for head in heads:
|
| 67 |
+
output = head['output']
|
| 68 |
+
batch_size, grid_h, grid_w, channels = output.shape
|
| 69 |
+
stride = head['stride']
|
| 70 |
+
|
| 71 |
+
bbox_part = output[:, :, :, :bbox_channels]
|
| 72 |
+
class_part = output[:, :, :, bbox_channels:]
|
| 73 |
+
|
| 74 |
+
bbox_part = bbox_part.reshape(batch_size, grid_h, grid_w, 4, reg_max)
|
| 75 |
+
bbox_part = bbox_part.reshape(grid_h * grid_w, 4, reg_max)
|
| 76 |
+
class_part = class_part.reshape(batch_size, grid_h * grid_w, class_channels)
|
| 77 |
+
|
| 78 |
+
for b in range(batch_size):
|
| 79 |
+
for i in range(grid_h * grid_w):
|
| 80 |
+
h = i // grid_w
|
| 81 |
+
w = i % grid_w
|
| 82 |
+
class_scores = class_part[b, i, :]
|
| 83 |
+
class_id = np.argmax(class_scores)
|
| 84 |
+
class_score = class_scores[class_id]
|
| 85 |
+
box_prob = sigmoid(class_score)
|
| 86 |
+
if box_prob < confidence_threshold:
|
| 87 |
+
continue
|
| 88 |
+
bbox = bbox_part[i, :, :]
|
| 89 |
+
dis_left = decode_distributions(bbox[0, :], reg_max)
|
| 90 |
+
dis_top = decode_distributions(bbox[1, :], reg_max)
|
| 91 |
+
dis_right = decode_distributions(bbox[2, :], reg_max)
|
| 92 |
+
dis_bottom = decode_distributions(bbox[3, :], reg_max)
|
| 93 |
+
pb_cx = (w + 0.5) * stride
|
| 94 |
+
pb_cy = (h + 0.5) * stride
|
| 95 |
+
x0 = pb_cx - dis_left * stride
|
| 96 |
+
y0 = pb_cy - dis_top * stride
|
| 97 |
+
x1 = pb_cx + dis_right * stride
|
| 98 |
+
y1 = pb_cy + dis_bottom * stride
|
| 99 |
+
scale_x = original_shape[1] / input_size[0]
|
| 100 |
+
scale_y = original_shape[0] / input_size[1]
|
| 101 |
+
x0 = np.clip(x0 * scale_x, 0, original_shape[1] - 1)
|
| 102 |
+
y0 = np.clip(y0 * scale_y, 0, original_shape[0] - 1)
|
| 103 |
+
x1 = np.clip(x1 * scale_x, 0, original_shape[1] - 1)
|
| 104 |
+
y1 = np.clip(y1 * scale_y, 0, original_shape[0] - 1)
|
| 105 |
+
width = x1 - x0
|
| 106 |
+
height = y1 - y0
|
| 107 |
+
detections.append(Object(
|
| 108 |
+
bbox=[float(x0), float(y0), float(width), float(height)],
|
| 109 |
+
label=int(class_id),
|
| 110 |
+
prob=float(box_prob)
|
| 111 |
+
))
|
| 112 |
+
|
| 113 |
+
if len(detections) == 0:
|
| 114 |
+
return []
|
| 115 |
+
boxes = np.array([d.bbox for d in detections])
|
| 116 |
+
scores = np.array([d.prob for d in detections])
|
| 117 |
+
class_ids = np.array([d.label for d in detections])
|
| 118 |
+
|
| 119 |
+
final_detections = []
|
| 120 |
+
unique_classes = np.unique(class_ids)
|
| 121 |
+
for cls in unique_classes:
|
| 122 |
+
idxs = np.where(class_ids == cls)[0]
|
| 123 |
+
cls_boxes = boxes[idxs]
|
| 124 |
+
cls_scores = scores[idxs]
|
| 125 |
+
x1_cls = cls_boxes[:, 0]
|
| 126 |
+
y1_cls = cls_boxes[:, 1]
|
| 127 |
+
x2_cls = cls_boxes[:, 0] + cls_boxes[:, 2]
|
| 128 |
+
y2_cls = cls_boxes[:, 1] + cls_boxes[:, 3]
|
| 129 |
+
areas = (x2_cls - x1_cls) * (y2_cls - y1_cls)
|
| 130 |
+
order = cls_scores.argsort()[::-1]
|
| 131 |
+
keep = []
|
| 132 |
+
while order.size > 0:
|
| 133 |
+
i = order[0]
|
| 134 |
+
keep.append(i)
|
| 135 |
+
if order.size == 1:
|
| 136 |
+
break
|
| 137 |
+
xx1 = np.maximum(x1_cls[i], x1_cls[order[1:]])
|
| 138 |
+
yy1 = np.maximum(y1_cls[i], y1_cls[order[1:]])
|
| 139 |
+
xx2 = np.minimum(x2_cls[i], x2_cls[order[1:]])
|
| 140 |
+
yy2 = np.minimum(y2_cls[i], y2_cls[order[1:]])
|
| 141 |
+
w = np.maximum(0, xx2 - xx1)
|
| 142 |
+
h = np.maximum(0, yy2 - yy1)
|
| 143 |
+
intersection = w * h
|
| 144 |
+
iou = intersection / (areas[i] + areas[order[1:]] - intersection)
|
| 145 |
+
inds = np.where(iou <= nms_threshold)[0]
|
| 146 |
+
order = order[inds + 1]
|
| 147 |
+
for idx in keep:
|
| 148 |
+
final_detections.append(Object(
|
| 149 |
+
bbox=cls_boxes[idx].tolist(),
|
| 150 |
+
label=int(cls),
|
| 151 |
+
prob=float(cls_scores[idx])
|
| 152 |
+
))
|
| 153 |
+
return final_detections
|
| 154 |
+
|
| 155 |
+
def main():
|
| 156 |
+
parser = argparse.ArgumentParser(description="YOLO11 AXEngine Inference")
|
| 157 |
+
parser.add_argument('--model', type=str, default='yolo11x.axmodel', help='Model path')
|
| 158 |
+
parser.add_argument('--image', type=str, default='dog.jpg', help='Image path')
|
| 159 |
+
parser.add_argument('--conf', type=float, default=0.45, help='Confidence threshold')
|
| 160 |
+
parser.add_argument('--nms', type=float, default=0.45, help='NMS threshold')
|
| 161 |
+
parser.add_argument('--size', type=int, nargs=2, default=[640, 640], help='Input size W H')
|
| 162 |
+
parser.add_argument('--regmax', type=int, default=16, help='DFL reg_max value')
|
| 163 |
+
args = parser.parse_args()
|
| 164 |
+
|
| 165 |
+
try:
|
| 166 |
+
input_tensor, original_shape, original_image = preprocess(args.image, tuple(args.size))
|
| 167 |
+
except FileNotFoundError as e:
|
| 168 |
+
print(e)
|
| 169 |
+
return
|
| 170 |
+
|
| 171 |
+
try:
|
| 172 |
+
session = axe.InferenceSession(args.model)
|
| 173 |
+
except Exception as e:
|
| 174 |
+
print(f"Error loading model: {e}")
|
| 175 |
+
return
|
| 176 |
+
|
| 177 |
+
input_name = session.get_inputs()[0].name
|
| 178 |
+
output_names = [output.name for output in session.get_outputs()]
|
| 179 |
+
|
| 180 |
+
try:
|
| 181 |
+
outputs = session.run(output_names, {input_name: input_tensor})
|
| 182 |
+
except Exception as e:
|
| 183 |
+
print(f"Error during inference: {e}")
|
| 184 |
+
return
|
| 185 |
+
|
| 186 |
+
try:
|
| 187 |
+
detections = postprocess(
|
| 188 |
+
outputs,
|
| 189 |
+
original_shape,
|
| 190 |
+
tuple(args.size),
|
| 191 |
+
args.conf,
|
| 192 |
+
args.nms,
|
| 193 |
+
reg_max=args.regmax
|
| 194 |
+
)
|
| 195 |
+
except Exception as e:
|
| 196 |
+
print(f"Error during post-processing: {e}")
|
| 197 |
+
return
|
| 198 |
+
|
| 199 |
+
for det in detections:
|
| 200 |
+
bbox = det.bbox
|
| 201 |
+
score = det.prob
|
| 202 |
+
class_id = det.label
|
| 203 |
+
if class_id >= len(COCO_CLASSES):
|
| 204 |
+
label = f"cls{class_id}: {score:.2f}"
|
| 205 |
+
else:
|
| 206 |
+
label = f"{COCO_CLASSES[class_id]}: {score:.2f}"
|
| 207 |
+
x, y, w, h = map(int, bbox)
|
| 208 |
+
cv2.rectangle(original_image, (x, y), (x + w, y + h), (0, 255, 0), 2)
|
| 209 |
+
cv2.putText(original_image, label, (x, y - 10),
|
| 210 |
+
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
|
| 211 |
+
|
| 212 |
+
cv2.imwrite('detections.png', cv2.cvtColor(original_image, cv2.COLOR_RGB2BGR))
|
| 213 |
+
print("结果已保存到 detections.png")
|
| 214 |
+
|
| 215 |
+
if __name__ == '__main__':
|
| 216 |
+
main()
|
yolo11_config.json
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_type": "ONNX",
|
| 3 |
+
"npu_mode": "NPU1",
|
| 4 |
+
"quant": {
|
| 5 |
+
"input_configs": [
|
| 6 |
+
{
|
| 7 |
+
"tensor_name": "images",
|
| 8 |
+
"calibration_dataset": "./dataset/coco_1000.tar",
|
| 9 |
+
"calibration_size": 32,
|
| 10 |
+
"calibration_mean": [0, 0, 0],
|
| 11 |
+
"calibration_std": [255.0, 255.0, 255.0]
|
| 12 |
+
}
|
| 13 |
+
],
|
| 14 |
+
"calibration_method": "MinMax",
|
| 15 |
+
"precision_analysis": true,
|
| 16 |
+
"precision_analysis_method":"EndToEnd"
|
| 17 |
+
},
|
| 18 |
+
"input_processors": [
|
| 19 |
+
{
|
| 20 |
+
"tensor_name": "images",
|
| 21 |
+
"tensor_format": "BGR",
|
| 22 |
+
"src_format": "BGR",
|
| 23 |
+
"src_dtype": "U8",
|
| 24 |
+
"src_layout": "NHWC"
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"output_processors": [
|
| 28 |
+
{
|
| 29 |
+
"tensor_name": "/model.23/Concat_output_0",
|
| 30 |
+
"dst_perm": [0, 2, 3, 1]
|
| 31 |
+
}, {
|
| 32 |
+
"tensor_name": "/model.23/Concat_1_output_0",
|
| 33 |
+
"dst_perm": [0, 2, 3, 1]
|
| 34 |
+
}, {
|
| 35 |
+
"tensor_name": "/model.23/Concat_2_output_0",
|
| 36 |
+
"dst_perm": [0, 2, 3, 1]
|
| 37 |
+
}
|
| 38 |
+
],
|
| 39 |
+
"compiler": {
|
| 40 |
+
"check": 2
|
| 41 |
+
}
|
| 42 |
+
}
|
yolo11_out.jpg
ADDED
|
Git LFS Details
|
yolo11s-cut.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:dd483e87031d8c7bd04aaf18e6a50cd71006c9b086d09df4eb3614cd6cab3bf0
|
| 3 |
+
size 37856530
|