View the ipython notebook with all of the code here Resultsįor each picture the blue dot is where the camera was, and the red dot is where the gaze was predicted to be. Since the network outputs the distance from the camera in cm, I could convert this to pixel space on the screen based on the resolution of the pictures and the physical dimensions of the phone. I then loaded their model, and fed the inputs through the network.įor all pictures tested I used the selfie camera in portrait mode on an IPhone 8. I used OpenCV to extract the face and eye features, and calculated the 25x25 face grid with code from faceGridFromRect.m that I converted to python.
The convolution neural network architecture. I struggled for days but finally setup Ubuntu with caffe and cuda on my macbook pro, which has a basic NVIDIA GeForce GT 650M graphics card. The ultimate deployment would be on the NVidia Jetson TX2, a single-board computer with a powerful GPU and support for six camera inputs.įor the neural network framework I stuck with caffe since that’s what the model is published in. I wanted to get it working in Python so that it would be platform agnostic the idea would be to eventually get this working on any computer with a webcam and decent GPU. My next goal for Presence was to try to get eye gaze prediction working using the Eye Tracking for Everyone model. View the ipython notebook with the code here Predicting Gaze in Python with the Eye Tracking for Everyone Neural Network Dan Oved's Blog ←Home Designing for digital fabrication Intro to physical computing Visual language Electronic rituals oracles and fortune telling Homemade hardware Energy Creating with interactive media Drawing on everything Nime Artists in the archive Basic analog circuits Haptics Soft sensing The temporary expert Tags.
0 kommentar(er)
