lvsn
about prediction and evaluation
Hi,
I have a question related to the model evaluation.
According to the repository, I must provide two csv files. Are these two files generated during the prediction stage? I train the tracker using the code in this repo, but I can't find the code for prediction.
I would like to know if there is any way you can help me with that information?
Thanks,
I found some npy files in the sequence folders, you can change a little bit of the code in the evaluation scripts and run them using the npy files
Hi,
That's right. I also found some npy files in the sequence folders: poses.npy and vicon2model.npy.
I think that the poses.npy file could be converted to the ground_truth.csv file. And, the prediction_pose.csv file is missing, because, as I understand, this must be generated with the prediction code. Right?
It is exactly the portion of code that I'm looking for in this repository, the one that makes the model evaluation and generate the results file! This repository contains the code for training the tracker, but does it have the code for evaluation? Not sure if I'm understanding the things right!
Thank you for your reply!
Hi,
I think vicon2model.npy is the ground truth, they mention in the paper that they use a vicon mocap system for ground truth collection, poses.npy could be the predicted poses. If you trained the tracker then you might have to write your own prediction code. I couldn't find that either.
As per my understanding, the evaluation codes use already predicted poses, but there no mention of exactly which network they used to get the prediction saved in the poses.npy files.
Hope i could help.
Ok. I'll try to implement the prediction code. This project is quite big and I don't want to invest lot of time coding only for benchmark proposes.
Thank you for your reply!
Hey, again sorry for the delay I was out of office... So I will link to an old thread about prediction #9 . I have some code that might need slight modifications on another repo that you could start from. I am quite busy these time but I will try to transfer my code to this repo when I have some cycles.