Closed alpertungakin closed 4 years ago
I'm not ML expert, but I take it that tiny-yolov2 has different architecture from mobilenet ssd, so it's expected that it's not gonna work. I imagine that for your model you'll have to modify DetectAsync methon in Detector.cs, figure out inputs/outputs and how to get boxes correctly.
Thank you. I will try and write the result here.
For YOLO model, you might want to check my new example: https://github.com/Syn-McJ/TFClassify-Unity-Barracuda
Hello, I was trying to use your project with tiny-yolov2 graph and coco labels. I trained the graph with tensorflow-gpu 1.4.0. I've not seen any error message but it cannot detect any object with Android phone. There wasnt any problem with other graphs. Here is the screenshot of inspector