Startup NPU into VIM3

VIM3
1

hello, I have lot of problem. I insert the commands below. How the whl files are not supported by the platform whereas it is the commands requested in the platform?

(venv) root@Khadas:/npu/npu# wget https://dl.khadas.com/Tools/TensorFlow/onnx-1.4.1-cp36-cp36m-linux_aarch64.whl
–2020-10-30 07:15:28-- https://dl.khadas.com/Tools/TensorFlow/onnx-1.4.1-cp36-cp36m-linux_aarch64.whl
Resolving dl.khadas.com (dl.khadas.com)… 64:ff9b::36c5:c5a5, 54.197.197.165
Connecting to dl.khadas.com (dl.khadas.com)|64:ff9b::36c5:c5a5|:443… connected.
HTTP request sent, awaiting response… 200 OK
Length: 3586846 (3.4M)
Saving to: ‘onnx-1.4.1-cp36-cp36m-linux_aarch64.whl’

onnx-1.4.1-cp36-cp36m-linux_aarch 100%[===========================================================>] 3.42M 1.47MB/s in 2.3s

2020-10-30 07:15:31 (1.47 MB/s) - ‘onnx-1.4.1-cp36-cp36m-linux_aarch64.whl’ saved [3586846/3586846]

(venv) root@Khadas:/npu/npu# wget https://dl.khadas.com/Tools/TensorFlow/scipy-1.2.0-cp36-cp36m-linux_aarch64.whl
–2020-10-30 07:15:50-- https://dl.khadas.com/Tools/TensorFlow/scipy-1.2.0-cp36-cp36m-linux_aarch64.whl
Resolving dl.khadas.com (dl.khadas.com)… 64:ff9b::36c5:c5a5, 54.197.197.165
Connecting to dl.khadas.com (dl.khadas.com)|64:ff9b::36c5:c5a5|:443… connected.
HTTP request sent, awaiting response… 200 OK
Length: 41569817 (40M)
Saving to: ‘scipy-1.2.0-cp36-cp36m-linux_aarch64.whl’

scipy-1.2.0-cp36-cp36m-linux_aarc 100%[===========================================================>] 39.64M 2.01MB/s in 20s

2020-10-30 07:16:11 (1.95 MB/s) - ‘scipy-1.2.0-cp36-cp36m-linux_aarch64.whl’ saved [41569817/41569817]

(venv) root@Khadas:/npu/npu# wget https://dl.khadas.com/Tools/TensorFlow/tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl
–2020-10-30 07:16:15-- https://dl.khadas.com/Tools/TensorFlow/tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl
Resolving dl.khadas.com (dl.khadas.com)… 64:ff9b::36c5:c5a5, 54.197.197.165
Connecting to dl.khadas.com (dl.khadas.com)|64:ff9b::36c5:c5a5|:443… connected.
HTTP request sent, awaiting response… 200 OK
Length: 126698057 (121M)
Saving to: ‘tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl’

tensorflow-1.10.1-cp36-cp36m-linu 100%[===========================================================>] 120.83M 1.90MB/s in 60s

2020-10-30 07:17:16 (2.00 MB/s) - ‘tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl’ saved [126698057/126698057]

(venv) root@Khadas:/npu/npu# pip install scipy-1.2.0-cp36-cp36m-linux_aarch64.whl
ERROR: scipy-1.2.0-cp36-cp36m-linux_aarch64.whl is not a supported wheel on this platform.
(venv) root@Khadas:/npu/npu# pip3 install scipy-1.2.0-cp36-cp36m-linux_aarch64.whl
ERROR: scipy-1.2.0-cp36-cp36m-linux_aarch64.whl is not a supported wheel on this platform.
(venv) root@Khadas:/npu/npu# uname -a
Linux Khadas 4.9.232 #2 SMP PREEMPT Sun Oct 25 21:10:43 CET 2020 aarch64 aarch64 aarch64 GNU/Linux
(venv) root@Khadas:/npu/npu# pip install onnx-1.4.1-cp36-cp36m-linux_aarch64.whl
ERROR: onnx-1.4.1-cp36-cp36m-linux_aarch64.whl is not a supported wheel on this platform.
(venv) root@Khadas:/npu/npu# pip install tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl
ERROR: tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl is not a supported wheel on this platform.
(venv) root@Khadas:/npu/npu# pip install keras==2.0
Collecting keras==2.0
Downloading Keras-2.0.0.tar.gz (191 kB)
|████████████████████████████████| 191 kB 925 kB/s
ERROR: Could not find a version that satisfies the requirement tensorflow (from keras==2.0) (from versions: none)
ERROR: No matching distribution found for tensorflow (from keras==2.0)

2

are you trying to get Kera python library for the NPU ?, if so please ask @Frank, as I don’t think its supported yet…

3

@ronron You can’t install keras in VIM3 . You should train in PC , then convert it and run with VIM3.

4

I had found the URL: How to install TensorFlow && keras with Ubuntu18.04
to begin.

Actually, I would like to detect objects on files by tensorflow and later do face detection :slight_smile:

5

@ronron Yes, this is the only TF version that can run on the board, but there is no keras version that can run . If you run TF directly on the board, you cannot use the NPU, which means you are using the CPU to run TF, which is not a good practice, so this cannot be compared with the PC, and the operation is very slow.

6

Yes, TF with CPU, it’s not good :slight_smile:. I managed to get the NPU to work :slight_smile:

7

@ronron Using NPU is the best way on VIM3