search

OlafenwaMoses / ImageAI

A python library built to empower developers to build applications and systems with self-contained Computer Vision capabilities
https://www.genxr.co/#products
MIT License
8.48k stars 2.18k forks source link

Can ImageAI use newer Libraries? #761

Open damies13 opened 2 years ago

damies13 commented 2 years ago

Hi I was wanting to try ImageAI but was not able to get it working following the Install Documentation

pip install tensorflow==2.4.0

The first issue was tensorflow 2.4.0 is no longer available so I just installed the latest version (2.9.1), I also tried installing the oldest available version 2.5.0

I also had the same issue with

pip install keras==2.4.3 numpy==1.19.3 pillow==7.0.0 scipy==1.4.1 h5py==2.10.0 matplotlib==3.3.2 opencv-python keras-resnet==0.2.0

For some of these libraries these version weren't available, I started as with tensorflow with the current version, then when I rolled back to tensorflow 2.5.0 I rolled these back to the closest available version

For Reference I'm using a vanilla Debian 11 install with Python 3.9.2

Other issues I had (that might be worth including in the documentation):

  1. The dependancy on libGL.so.1 was missed, so I needed to figure out that I needed to install 
    apt install libgl1-mesa-glx
  2. tensorflow needs a lot of memory to install (>2GB), for my application I intend to use small ARM SBC's these often only have 512MB of ram so I set my VM up with the same specifications, however the tensorflow install just crashed, I needed to give more than 2GB of swap memory in order for tensorflow to install at all.

My main question is: Will there be a version soon of ImageAI that works with a more recent version of tensorflow and the other dependancies?

Dave.

alonsoir commented 1 year ago

This issue is quite important because right now anyone who doesn't have the dependencies saved by chance on their computer won't be able to configure the framework. I understand the difficulty of maintaining the software with the new versions of the dependencies, but it is absolutely necessary to keep the project alive.

YakFromStateFarm commented 1 year ago

I agree with the above and am willing to offer any help needed if that is of any interest.

immortal-blog commented 1 year ago

I agree. I cannot install and run directly because it uses too many old libs that don't exist on pip. I don't want to find the old versions and install them one by one. I just give up.

tankxiaodi commented 1 year ago

You might need to check the tensorflow-2.7.2 branch.