got poor result using my mobilenetssd model

Syn-McJ / TFClassify-Unity

An example of using Tensorflow with Unity for image classification and object detection.

MIT License

167 stars 47 forks source link

got poor result using my mobilenetssd model #38

Open WisionChan opened 4 years ago

WisionChan commented 4 years ago

Hi, i'm new in Unity and try to use my hand detect model in Unity, and this project seems very good for me,thanks~~

After replacing the model and label with my own model(trained by tensorflow 1.7.1+MobilenetSSD),the result is quite poor, only about 0.003,which i can achieve 0.98 or higher in Android or python.

So i wonder whether should i change something or some code like IMAGE_STD or somewhat?

Any reply will be appreciated~

Syn-McJ commented 4 years ago

Hi @WisionChan, what do you mean by result of 0.003, is that confidence level?

In general, I'm not sure how to improve this example further. One thing might be worth trying is to use AR camera from ARFoundation and try to get a snapshot from it. I think that the image that I'm getting from WebCamTexture might be of poor quality.

syedali1621 commented 3 years ago

@WisionChan which repository you used to train your mobilenetSSD model?