Update README.md
Browse files
README.md
CHANGED
@@ -1,6 +1,109 @@
|
|
1 |
---
|
2 |
license: bsd-3-clause
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
4 |
|
5 |
-
|
6 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: bsd-3-clause
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
base_model:
|
6 |
+
- depth-anything/Depth-Anything-V2-Small
|
7 |
+
pipeline_tag: depth-estimation
|
8 |
+
tags:
|
9 |
+
- Depth-Anything-V2
|
10 |
---
|
11 |
|
12 |
+
# Depth-Anything-V2
|
13 |
+
|
14 |
+
This version of Depth-Anything-V2 has been converted to run on the Axera NPU using **w8a16** quantization.
|
15 |
+
|
16 |
+
This model has been optimized with the following LoRA:
|
17 |
+
|
18 |
+
Compatible with Pulsar2 version: 3.4
|
19 |
+
|
20 |
+
## Convert tools links:
|
21 |
+
|
22 |
+
For those who are interested in model conversion, you can try to export axmodel through
|
23 |
+
|
24 |
+
- [The repo of original](https://depth-anything-v2.github.io/)
|
25 |
+
|
26 |
+
- [The repo of AXera Platform](https://github.com/AXERA-TECH/DepthAnythingV2.axera), which you can get the detial of guide
|
27 |
+
|
28 |
+
- [Pulsar2 Link, How to Convert ONNX to axmodel](https://pulsar2-docs.readthedocs.io/en/latest/pulsar2/introduction.html)
|
29 |
+
|
30 |
+
|
31 |
+
## Support Platform
|
32 |
+
|
33 |
+
- AX650
|
34 |
+
- [M4N-Dock(爱芯派Pro)](https://wiki.sipeed.com/hardware/zh/maixIV/m4ndock/m4ndock.html)
|
35 |
+
- [M.2 Accelerator card](https://axcl-docs.readthedocs.io/zh-cn/latest/doc_guide_hardware.html)
|
36 |
+
- AX630C
|
37 |
+
- [爱芯派2](https://axera-pi-2-docs-cn.readthedocs.io/zh-cn/latest/index.html)
|
38 |
+
- [Module-LLM](https://docs.m5stack.com/zh_CN/module/Module-LLM)
|
39 |
+
- [LLM630 Compute Kit](https://docs.m5stack.com/zh_CN/core/LLM630%20Compute%20Kit)
|
40 |
+
|
41 |
+
|Chips|Time|
|
42 |
+
|--|--|
|
43 |
+
|AX650| 33 ms |
|
44 |
+
|AX630C| 310 ms |
|
45 |
+
|
46 |
+
## How to use
|
47 |
+
|
48 |
+
Download all files from this repository to the device
|
49 |
+
|
50 |
+
```
|
51 |
+
root@ax650:~/AXERA-TECH/Depth-Anything-V2# tree
|
52 |
+
.
|
53 |
+
|-- README.md
|
54 |
+
|-- calib-cocotest2017.tar
|
55 |
+
|-- config.json
|
56 |
+
|-- depth_anything_v2_vits.onnx
|
57 |
+
|-- depth_anything_v2_vits_ax620e.axmodel
|
58 |
+
|-- depth_anything_v2_vits_ax650.axmodel
|
59 |
+
|-- examples
|
60 |
+
| |-- demo01.jpg
|
61 |
+
....
|
62 |
+
| `-- demo20.jpg
|
63 |
+
|-- output-ax.png
|
64 |
+
`-- python
|
65 |
+
|-- infer.py
|
66 |
+
|-- infer_onnx.py
|
67 |
+
|-- output.png
|
68 |
+
`-- requirements.txt
|
69 |
+
|
70 |
+
2 directories, 31 files
|
71 |
+
root@ax650:~/AXERA-TECH/Depth-Anything-V2#
|
72 |
+
```
|
73 |
+
|
74 |
+
### python env requirement
|
75 |
+
|
76 |
+
#### pyaxengine
|
77 |
+
|
78 |
+
https://github.com/AXERA-TECH/pyaxengine
|
79 |
+
|
80 |
+
```
|
81 |
+
wget https://github.com/AXERA-TECH/pyaxengine/releases/download/0.1.3.rc1/axengine-0.1.3-py3-none-any.whl
|
82 |
+
pip install axengine-0.1.3-py3-none-any.whl
|
83 |
+
```
|
84 |
+
|
85 |
+
#### others
|
86 |
+
|
87 |
+
Maybe None.
|
88 |
+
|
89 |
+
#### Inference with AX650 Host, such as M4N-Dock(爱芯派Pro)
|
90 |
+
|
91 |
+
Input image:
|
92 |
+
|
93 |
+

|
94 |
+
|
95 |
+
```
|
96 |
+
root@ax650:~/AXERA-TECH/Depth-Anything-V2# python3 python/infer.py --model depth_anything_v2_vits_ax650.axmodel --img examples/demo01.jpg
|
97 |
+
[INFO] Available providers: ['AxEngineExecutionProvider']
|
98 |
+
[INFO] Using provider: AxEngineExecutionProvider
|
99 |
+
[INFO] Chip type: ChipType.MC50
|
100 |
+
[INFO] VNPU type: VNPUType.DISABLED
|
101 |
+
[INFO] Engine version: 2.12.0s
|
102 |
+
[INFO] Model type: 2 (triple core)
|
103 |
+
[INFO] Compiler version: 3.3 ae03a08f
|
104 |
+
root@ax650:~/AXERA-TECH/Depth-Anything-V2# ls
|
105 |
+
```
|
106 |
+
|
107 |
+
Output image:
|
108 |
+
|
109 |
+

|