Systems and methods for 3D image distification
摘要
說明書
At block 708 the computing device can then generate a prediction model, for example, by training the model using the output feature vector(s). In certain embodiments, the prediction model is trained using a neural network, such as a convolutional neural network as described herein. In other aspects, the prediction model may be trained with batches of the output feature vector(s), where batches of the output feature vector(s) correspond with subset(s) of 3D images taken from the set of the one or more 3D images. Thus, for example, a set of 67,000 images may be batched into 1000 images subsets used to trained the model.
In other aspects, the 3D images of the prediction model of method 700 may relate to driver behaviors as described herein. After the model method 700 is generated, it may be used to classify 3D images. For example, a second set of 3D images may be input into the predictive model and used to determine, for each 3D image of the second set, a driver behavior classification and a corresponding probability value that indicates the probability that the 3D image is associated with the driver behavior classification. In some aspects, the classification with the highest probability value determines the overall image classification for the 3D image. In other aspects, multiple classifications and corresponding values are maintained for each of the 3D images.
In some embodiments, for some of the 3D images in the second set, the driver behavior classification and the probability value can be transmitted to a different computing device, such as remote computing device or any other device described for
