/humanProb

A machine learning model to predict the probability that a person is indeed a human in a given video.

Primary LanguageJupyter Notebook

A Recurrent Neural Net that returns a list of probabilities that whether the given element, keyed by ID according to the openpose results are actual human or not. Where a probability of more that 0.9 denotes it is a human for sure and a probability of less than 0.1 denotes that it is not.

OUTPUT FORMAT

Here, the output is of the ndjson format. It is the OpenPose's COCO Pose format in the form of JSON file. Each of the reading has an x-coordinate, y-coordinate and the probability of the prediction, i.e. [x,y, p(x,y)]. To know more about the OpenPose output format, refer: https://github.com/CMU-Perceptual-Computing-Lab/openpose/

Requirements:

Tensorflow-GPU 2.0
CUDA (Nvidia GPU)