I trained a RetinaNet to classify 5 LEGOs. You can find the weights here. Here are some Programs which might help you train and using a RetinaNet or a YOLOv2. For this I used the "keras-retinanet" Library. It is essential that you install it properly before you use the program "Website". I used Node.js V8.12.0 and Python V3.5.0.
- Fotostation - Creates a server. You can take Fotos with any device and it will be uploaded on this server. You NEED to make your own Server Certificates! Here is a Turorial. Tested on Android.
- Website-Dev - When you place Images in the imgs/imgsToClassifiy folder, you can annotate them via Browser
- GreenScreen - Generate an artificial Dataset. You need png Images with transparent Background and images as Backgrounds
- toCSV - Translate .xml annotations into one .csv annotation
- Website - If configured properly (classify.py got copied in the keras-retinanet-master folder, the weights got downloaded and copied in the keras-retinanet-master/snapshots folder and in routs.js and classify.py the Path got changed) you can visit this Website, upload your Image and it will be classified.
- Install Node.js V8.12.0 and keras-retinanet (Python 3.5.0)
- Every Node.js program-folder run ´npm install´
- If you want to launch Fotostation, Website or Website-Dev use the ´node server.js´ command
After finishing the previous steps you need to make a certificate. Make it using openssl. Paste the Certificates in the Folder and make sure that they are named the same as referenced in the server.js file.
Firstly you need to install keras-retinanet. If you have installed it properly just paste in the keras-retinanet-master folder the classify.py script. You will NEED to edit the file. You will need to specify ´os.chdir´. Reference where your keras-retinanet-master folder is located. Then you NEED to edit the routs.js file. You must edit the shell variable. Also there reference you keras-retinanet-master folder.