Problem about load_hair

[Hacetate]

When I load the .data using io,py,I encountered this error: strands = np.array(strands).reshape((num_strands, -1,3)) ValueError: cannot reshape array of size 10000 into shape (10000,newaxis,3) .
The reason for this is that numpy reads an irregular list and append to strands list, which cannot convert to a reshapable Numpy Array.
So what I need to do is zeroize or something else?

[c-he]

Which .data file are you trying to load? Its size 10000 seems a bit weird to me.

[Hacetate]

The Raw .data of USC-Hairsalon dataset all have strand of around 10,000

[c-he]

Oh I see, the original data you downloaded from their website right? These data contain some irregular strands which only have one point. You can download our processed data from this OneDrive link. We filtered those strands and aligned them to our head mesh.

[Hacetate]

Thanks for the resampled original data. Could you please provide the code for preprocessing irregular data? I found that the dataset of the original USC-hairsalon is all 10000x300. However, the data from Hairnet interpolation contains a large number of strands greater than 300, while others range from 100 to 200. To obtain a better training dataset, it should not be sufficient to simply truncate strand above 300 and delete strand less than 300.

[c-he]

Sorry for the late reply, do you mean the data you downloaded from Hair20k cannot be properly loaded by load_hair()? If so, could you provide some sample data? We can look into it later.

[wrk226]

Sorry for the late reply, do you mean the data you downloaded from Hair20k cannot be properly loaded by load_hair()? If so, could you provide some sample data? We can look into it later.

In fact, most hair data cannot be successfully loaded using load_hair(), because each strand has a different length. That is, the imported strands are a list of lists with unequal lengths. For example, you can check the file:

full_placed\hair_placed\L_curly\data\head\strands00009_00026_10001.data

In this file, the first strand has 381 points while the second strand has 117 points. When running:

strands = np.array(strands).reshape((num_strands, -1, 3))

an error is raised:

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (9936,) + inhomogeneous part.

[c-he]

Can you try the Hair20k data downloaded form this link: https://yaleedu-my.sharepoint.com/:f:/g/personal/chengan_he_yale_edu/EvTqlju7SolPp8vJMgAqczABEAJKQBSt3HBN_Yn8WogOIg? I assume you are using the data downloaded from the Hair20K website, which may not be fully proprocessed.

[wrk226]

You're right. The data downloaded from the new link you sent has no issues at all. Thank you very much.