xrenaa / human-motion-analysis-with-deep-metric-learning

pytorch implement of this paper: https://arxiv.org/abs/1807.11176

Hi there 👋

human-motion-analysis-with-deep-metric-learning's Issues

FileNotFoundError: [Errno 2] No such file or directory: './log/model_new_4999.pth'

Does anyone have this problem? What's the reason.
FileNotFoundError: [Errno 2] No such file or directory: './log/model_new_4999.pth'

How do you test for action recognition?

What is the strategy during test step? Are you doing the forward propagation for all test sequences to extract the test features? Then how exactly to do the classification for action recognition problem? Thank you.

loss function

Hello, in the code, L2_Norm changes the 175128-dimensional vector to a 175-dimensional vector, and then the view becomes 725, which contains X, X+. Why?

How can I use the generated matchmaker to generate a gif of the character again?

Hello author, now I have got the generated matchmaker, but I don't know how to use the matchmaker to generate the corresponding character gifs, does the code you uploaded include this part? Or do I need to go to everyone dance now to download the code to generate the corresponding gifs?

the data

The data you provided on drive is lost.
Could you share it again?or give more information about the data format.
you said the key of the json file is "0","1",.....,"15",Each key contains:( _ , 50, 2, 17) pose.what is the 50 mean? And in the another issue,you said the 2 channels are the x and y coordinates, is that mean the value of the 2 channels are like
[743.7080688476562, 223.71688842773438] and the 17 mean the keypoint like "left_elbow"
or 2 channels are just the character x and y, and the 17 is the 17 coordinates in x?

i learn the Deep Learning just a little time, and im in great request of your help
Thank you in advance

pretrain model

Since the data is too large, can you provide a pretrained model for testing? Thanks

How to use this model to evaluate the performance of "Self-supervised Dance Video Synthesis Conditioned on Music"?

Hi,
I tried to reproduce the evaluation result in your paper "Music-Oriented Dance Video Synthesis with Pose Perceptual Loss".
Yet, I do not understand how to get the cross-modal evaluation numbers from this implementation.
I have finished training this model and the dance generative model.
My question is that how should I input the skeleton sequence generated from "Self-supervised Dance Video Synthesis Conditioned on Music" into this cross-modal metric model, and acquire the scores?

Thank you in advance!

How to normalize each pose?

How to use this model to measure sequence similarity

Hello,please I have some questions：
1、you said that each key contains:( _ , 50, 2, 17) pose.what is the 50 mean?I understand it is the length of each piece of data, which is related to the sampling rate. According to your other paper, it is about 5 seconds.
2、How can I use this model to measure sequence similarity? Is it a direct input of two sequences X and Y. And is the value of L2 norm directly returned and then mapped to a score？
Thank you for advance!

data

The data you provided on drive is lost.
Could you share it again?

Data format

I was wondering a bit about how to interpret the data format. Am I correct in assuming that the two channels are x and y coordinates for the 17 pose keypoints, with the first presumably being the x coordinates?

Thank you in advance