search

doughtmw / HoloLens2-Machine-Learning

Using deep learning models for image classification directly on the HoloLens 2.
MIT License
108 stars 28 forks source link

Video Frame settings question. #3

Closed dngkatn closed 2 years ago

dngkatn commented 3 years ago

Dear. Michell Doughty. Hello. I ran the sample you uploaded. I distributed it to Hololens 2, and it works very well. I just have a question. Is there a reason why you set the size of the video frame to (224, 224)? To stare at the image hololens for verification, I have to stick close to the monitor. Is this because of the size of the video frame? Can I increase the size of the video frame? I look forward to hearing from you. Thank you.

doughtmw commented 2 years ago

Hi @dngkatn, I stuck with 224 x 224 px video frames as they offer a good balance between model accuracy and inference speed. In this sample, I'm cropping video frames directly from their HoloLens 2 recorded resolution to 224 x 224 px for inference.

In another sample of mine [link], I include a C++ dll project (OpenCVRuntimeComponent) to replicate a typical image pre-processing pipeline from PyTorch training where the HoloLens 2 video frames are resized rather than cropped, resulting in much better performance.