Timeline for Making neural nets more efficient in Unity
Current License: CC BY-SA 4.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 5, 2018 at 16:22 | comment | added | Samuel Lopes | I believe TensorFlowSharp is .NET bindings for the c implementation of TensorFlow. I am loading a frozen .pb file into TensorFlowSharp and running it. | |
| Jun 5, 2018 at 16:10 | comment | added | starikcetin | Use the profiler to find out the bottlenecks and post them here, otherwise, we all would be just speculating. | |
| Jun 5, 2018 at 15:59 | comment | added | blurry | I'm not familiar with TensorFlowSharp, but it looks like a way to call the Python from C#? This could be a significant performance degradation as Python is (in general) far slower than compiled C#. If you can verify that you are calling into the Python itself I can give an answer as my job is literally a job about taking neural nets from python output and translating them into C# models. | |
| Jun 5, 2018 at 15:53 | comment | added | Samuel Lopes | It doesn't have to run on every frame, but this would be ideal | |
| Jun 5, 2018 at 15:53 | comment | added | Samuel Lopes | I am using this hand-object detection model. I believe it is SSD which means it is more lightweight than R-CNN's: github.com/victordibia/handtracking | |
| Jun 5, 2018 at 15:51 | comment | added | blurry | How large is your neural net, how many hidden layers do you have, and does it run on every frame? | |
| Jun 5, 2018 at 15:43 | history | asked | Samuel Lopes | CC BY-SA 4.0 |