Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Keypoint RCNN predicted probabilities greater than 1?

See original GitHub issue

I am performing inference using a pretrained model (keypoint_rcnn_R_101_FPN_3x). Given an input image, the keypoint output should be N x K x 3, where N is the number of detected persons, K=17 is the number of keypoints and each keypoint should be (x, y, prob) according to the documentation of function keypoint_rcnn_inference in detectron2/modeling/roi_heads/keypoint_head.py. However, some of the “probabilities” I get are above 1. Did I misunderstand something? Is this expected?

Issue Analytics

State:
Created 4 years ago
Reactions:3
Comments:7

Top GitHub Comments

1reaction

ppwwyyxxcommented, Feb 28, 2020

In theory there is, but I cannot help with it. I don’t think knowing a maximum is useful for anything. You can assume the maximum is 10 because I think the maximum is smaller than that.

1reaction

cm107commented, Feb 28, 2020

@ppwwyyxx Is there a way to calculate this maximum? If possible I would like rescale the score on a range of 0 to 1. Is there a good reason why it isn’t rescaled between 0 to 1 by default?

Top Results From Across the Web

Human Pose Estimation using Keypoint RCNN in PyTorch

The Region-Proposal-Layer predicts the rough location of N number of objects detected in the feature map. These variable-sized regions are then ...

Keypoint and Bounding Box Detection using PyTorch ...

In this tutorial, we are going to learn how we can detect keypoints and bounding boxes using Keypoint RCNN deep learning model and...

Keypoint RCNN visibility flag for keypoints - PyTorch Forums

I am fine-tuning Keypoint RCNN on my custom dataset and, ... Keypoint RCNN always predicts a flag value of v=1 for all keypoints, ......

Cascade Mask R-CNN and Keypoint Detection used in ...

The work in this thesis explores one new approach when used in combination with an instance segmentation model. The goal is to have...

Local Keypoint-based Faster R-CNN - Repository@Hull

Keywords - Keypoint; SIFT; Convolutional Neural Network; Faster R-CNN. 1. ... usually more computationally efficient than two-stage detectors [1][24].