Inference Tester is a web application that can test the inference of several DL pre-trained models on pollens. This app is a proof-of-concept developed to test the inferences on our models from beeopicture. We used Adobe XD to design the app and React/Express for the development.
It takes in input one or more images of pollens analyzed in the laboratory (x100), allows to choose between SSD, YOLOv5, EfficientDet and Faster-RCNN. The project comes with a YOLOv5 PyTorch implementation from Ultralytics. The notebook is in the beeopicture repository.
For the React client, you need to go to the client folder to install the required packages. You have to do the same for the Express server.
npm installAfter that, go to deepnet/YOLOv5 and unzip the zip file. Go inside the newly created folder (yolov5) and install the requirements.txt.
pip -r install requirements.txtYou have to make a few adjustements in the code to make it work. First in yolov5/detect.py, you need to replace the path by yours to the inferenceTester folder at the line 148.
Same in server/config/lib.js. Put your path at the line 7 and then whether you are using a conda env or not change the python command at line 21.
When the configuration is done, you just need to start the client with npm start and you can start the server with nodemon or node.
You can only test the inference with YOLOv5 on a CPU at the moment. We might add the support of the other algorithms on the list later on.
- YOLOv5
- EfficientDet-v2
- SSD
- Faster-RCNN