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

Boxes aren't drawn in Detect scene #31

Closed AnaRhisT94 closed 4 years ago

AnaRhisT94 commented 5 years ago

Hi Andrey, ( @Syn-McJ )

When I'm using Detect scene on my Android device, it doesn't drawn any rectangles on the screen.

However, with play button and using another camera, it sometimes draws the rectangle on the whole screen, but very not accurate. (I don't need it to work on the editor anyways)

Any ideas why this happens? I see in your README file that you did get the boxes drawn.

AnaRhisT94 commented 5 years ago

I always used Build & Run settings when first time I added the Classify scene. How I found it out? I went to Build settings and I noticed that it was building the Classify scene all the time. Changed it to Detect and problem solved.

Btw, the boxes that are drawn are mostly on the whole screen. Also, the detection is very slow, it takes a couple of seconds and detects just one thing. Any ideas how to solve it? What phone did you use? Maybe it about settings problem?

Syn-McJ commented 5 years ago

Hi @AnaRhisT94,

I see the problem with boxes, it relates to changes that happened in Unity which made camera texture have different size. I'll check how to fix this.

As for detection being slow, I mentioned this in readme and I still don't know how to improve this. Unity released their own inference library called Barracuda to replace TensorFlow Sharp, however last time I tried it I couldn't make it work with my model, but perhaps you'll have better luck.

Syn-McJ commented 4 years ago

I'm going to close this issue, feel free to reopen if the problem persists. You can also check my new Barracuda example, it should have better performace, but only supports YOLO for now.