About requirements.txt
cv1995 opened this issue · 5 comments
where is the requirements.txt?
Hi, sorry for late reply.
The requirements.txt is uploaded. However, there might be some errors during "pip install -r requirements.txt". If this happens, just installl the packages one by one. Sorry for the inconvenience.
Hi, sorry for late reply.
The requirements.txt is uploaded. However, there might be some errors during "pip install -r requirements.txt". If this happens, just installl the packages one by one. Sorry for the inconvenience.
Thank you for your reply. When I used your method to test in other scenes, I found that the performance was not very good, and some joints of the reconstructed 3D pose would be wrong. The fps of our video is 60. Here are the parameters I select.How should I adjust these parameters?
That's possible, since these parameters play an important role in this method. You have the find the exact parameters for different dataset, for example, those thresholds for 2D-3D association and joint filters need larger number when people are closer to the cameras. Or the n_init, max_age might relate to the fps, the higher fps, the larger (according to my experience).
That's possible, since these parameters play an important role in this method. You have the find the exact parameters for different dataset, for example, those thresholds for 2D-3D association and joint filters need larger number when people are closer to the cameras. Or the n_init, max_age might relate to the fps, the higher fps, the larger (according to my experience).
What do these two parameters (n_init, max_age) mean？How does it affect the final result?I don't know how to adjust them.Can I add your email for further communication?
That's possible, since these parameters play an important role in this method. You have the find the exact parameters for different dataset, for example, those thresholds for 2D-3D association and joint filters need larger number when people are closer to the cameras. Or the n_init, max_age might relate to the fps, the higher fps, the larger (according to my experience).
Hi, here are the concepts for these parameters, both the parameters are related to tracking:
- n_init : it represents that the 3D skeleton must keep matching with at least a 2D pose from a camera view for n iteration after initialized, since some 3D skeletons might be accidentally formed (false positive). So if an 3D skeleton pass this constraint, we consider is as a true positive。
- max_age: it stands for time threshold that a 3D skeleton doesn't associated to any 2D pose. So if a 3D skeleton haven't matched with any 2D pose longer than max_age. We consider the person leave the scene and delete it.
Thus, according to my experience, the higher the dataset's fps is, both the larger values to be set. However, I can't say which values are the best. You can try different numbers and find the best combination. When I was evaluating the result, it took me a while to find them.
This is my email address: b10532021@gapps.ntust.edu.tw
Feel free to contact if you have further questions.