search

stark-t / PAI

Pollination_Artificial_Intelligence
5 stars 1 forks source link

Building a simple app with a GUI that can run detection on a local PC (CPU & GPU) #63

Closed valentinitnelav closed 1 year ago

valentinitnelav commented 1 year ago

We discussed the idea of delivering a web page where people could upload their images, but it is unclear at the moment how can we do that and where can we store this site for free.

What interests me now is to develop a simply GUI based app that can be installed on a local PC from an executable file and then a user can simply upload an image or a folder with images and then the app make a system call to detec.py using the downloaded weights.

I think what will be tricky is to have all dependencies in place and most probably we will have to export the weights to something even leighter (a different format than the default one from YOLO).

valentinitnelav commented 1 year ago

At a fast search i just realized that there is a YOLOv5 app from ultralytics and one can run the models nano, small and medium. I tried it out from here https://ultralytics.com/yolov5 and YOLOv5-m moves also ok-sih on my Android phone, but nano feels to run inference really smooth.

Also there is this competition for deploying on edge devices organized by ultralytics https://github.com/ultralytics/yolov5/discussions/3213

stark-t commented 1 year ago

yolov5tflite

valentinitnelav commented 1 year ago

FYI - Just found out that YOLOv5 can be deployed as Python package now: https://pypi.org/project/yolov5/

valentinitnelav commented 1 year ago

I'll give up on this because of the legal implications that are not clear related to being able to share the weights (if they are considered derivative work based on the "GBIF" images", then we cannot share them, so I cannot put them on a server or on a GUI application).