Hello guys, I try to install object detection API tensorflow but tf-models-official library support with tensorflow 2.4.0, I try on windows it work. About jetson nano just publish tensorflow 2.3.1 version how can I fix this problem it.
You can build the package on your own.
Here is a sharing from the community for your reference:
The script is for version 2.3.0, but you can update it to 2.4.0 by replacing the source directly:
wget https://github.com/tensorflow/tensorflow/archive/v2.4.0.tar.gz -O tensorflow-2.4.0.tar.gz
I found object detection API tensorflow support with tensorflow 2.3.1 on jetson nano but We need build Tensorflow addons 2.3.0 , Apache Arrow form source. and dataclasses pip must install support with python3.6 ( version 0.7 ) newest version support with python 3.7. About avro-python3 need 1.8.1
- dataclassed need insatll verison 0.7 (support with python3.6 only)
- avro-python3 need install verison 1.8.1
- tf-models-official library need Tensorflow-addons while it’s didnt support with
- apache-beam library need Apache Arrow while it’s didn’s support with
- when we install lvis library request by object detction API tensorflow with pip
will install opencv-python 4.5.0 erorr with numpy
- tensorflow 2.3.1 need numpy <1,19,1
- opencv 4.5.0 need numpy > 1.19.1
How can we solve this problem with jetson nano ?
I really want build my own training object detection API with tensorflow.
I can run pre-trained models with jetson nano just install ‘Pillow>=1.0’, ‘Matplotlib>=2.1’, 'Cython>=0.28.1’and add path your models/research/object_detection to PYTHONPATH. but it can’t training my own object detection API tensorflow.
How about to downgrade your OpenCV version?
If you want to build OpenCV from the source, please check the below script.
You can change the source version by updating the
wget target directly.