search

zztant / MGANs

This repository contains code for the paper "Multi-task GANs for View-specific Feature Learning in Gait Recognition"
16 stars 4 forks source link

How to get the np files for training? #1

Open htZhang25 opened 5 years ago

htZhang25 commented 5 years ago

Hello,Thank you for your work.I quite don't know how to prepare my np files as you said"Please prepare your training data as np files"in Readme file. I'm confused is this np file just the .mat file which should be loaded when I run the gait.py?

zztant commented 5 years ago

Hello,Thank you for your work.I quite don't know how to prepare my np files as you said"Please prepare your training data as np files"in Readme file. I'm confused is this np file just the .mat file which should be loaded when I run the gait.py?

I am sorry for some negligence in the README, you only need to prepare the .mat file. For example in CASIA-B， 'data' is the PEIs of all sequences in the CASIA-B. 'view' is view information (1-11) of all sequences. 'label' is the identity information (1-124) of all sequences. 'cov' is the walking condition (1-10) of all sequences. The meaning of these numbers can be found in the .ipynb file.

htZhang25 commented 5 years ago

Sorry,I'm still confused about how to get the right .mat file. Should the .mat file contain all datas produced by different gait sequences folders (for example 124 subjects11 views 4 covs , just 5456 training examples)？Can you tell me some details about how to produce the .mat file ,such as what the dimention of view \label\ cov should be？I will be waiting for your reply sincerely.

zztant commented 5 years ago

Sorry,I'm still confused about how to get the right .mat file. Should the .mat file contain all datas produced by different gait sequences folders (for example 124 subjects11 views 4 covs , just 5456 training examples)？Can you tell me some details about how to produce the .mat file ,such as what the dimention of view \label\ cov should be？I will be waiting for your reply sincerely.

In the CASIA-B dataset, the number of sequence is about 124x11x10=13640. Therefore, the dimension of 'label', 'view' and 'cov' is [13640], and the dimension of 'data' is [13640, channel_num, height, width].

I'm sorry for making you confused. I didn't understand how to write code elegantly in deep learning project when I was doing this project. I strongly recommend you to rewrite the code of loading data by using the PyTorch API such as 'torch.utils.data.Dataset' and 'torch.utils.data.DataLoader'.