I currently use an object detector to detect an object and specific parts of it (a crop and its stem). Such detector is not the best choice for detecting parts that could be represented by a point (typically a stem) so I'm planning on moving to a keypoint detector.
After reading the literature it appears that there are many solutions. I'm particularly interested in using a hourglass network to predict a set of heatmaps abstracting the keypoints positions.
The problem is that many of the existing frameworks are dedicated to human pose estimation (for instance OpenPose) but I don't need all this complexity.
By best choice for now is this framework that is a Tensorflow implementation of Hourglass Networks but it is still too specific to human pose estimation.
Do you have any suggestion of frameworks that best suit my application? i.e. simple keypoint detection.