hkust-vgd/scenenn

The download link is broken

Opened this issue · 10 comments

chester256 commented

The link https://drive.google.com/drive/folders/0B-aa7y5Ox4eZWE8yMkRkNkU4Tk0 is broken, as well as the download script. Could you please update the link? Thanks.

songuke commented
chester256 commented

@songuke Thanks for your reply. The backup link works. However, I am confused about the label channel in the .ply file in main directory. I queried a sample point in 005.ply and it showed the label is 8844 (Shown in attached screenshots). What does 8844 mean?

The README in main directory says that "To parse this channel, please refer to the source code of our annotation tool". However, in the annotation tool release page, the source code zip file does not contain the source code. Could you please provide a reference for label channels in ply file? Thank you very much.

Screen Shot 2019-09-03 at 11 01 07

songuke commented
chester256 commented

Thanks for your reply. Actually we would like to do category-based semantic segmentation, but we want to sample points with stride and block size different from aforementioned preprocessed data. So we need to re-preprocess the category-based data.

Is the aforementioned preprocessed data (CVPR'18) from the category annotation? It seems that there are more samples in CVPR'18 data (up to scene id 700) than category annotation data (up to scene id 322).

Or should I re-preprocess the data from the instance segmentation? The labels in instance segmentation xml seem very few. Thank you very much.

songuke commented
chester256 commented

Thanks for your reply. However, there are only 50 scenes in category annotation (http://103.24.77.34:8080/scenenn/contrib/nyu_class/) while CVPR'18 preprocessed data provides 76 scenes. Is there 76 scenes category annotation (unpreprocessed) dataset link?

In addition, you mentioned that some categories are not available in the data, so how to deal with the unavailable categories? Thank you very much.

songuke commented
chester256 commented

Thanks! One more question, in the raw category annotation xml files, there are many points whose nyu_class is none, e.g. '<label id="283951" color="198 98 246" text="" nyu_class="" ...>', so should we label these points as unknown class or otherstructure/otherfurniture/otherprop classes? Thank you very much.

chester256 commented

Thanks! One more question, in the raw category annotation xml files, there are many points whose nyu_class is none, e.g. '<label id="283951" color="198 98 246" text="" nyu_class="" ...>', so should we label these points as unknown class or otherstructure/otherfurniture/otherprop classes? Thank you very much.

Problem solved. I found there is nyu_class field in the .ply file. Thanks again!

songuke commented

Thanks for finding this out. I should have the raw data of the 76 scenes. Let me find and get back to you in this thread by Wednesday night. 2019年9月3日(火) 15:57 zzy notifications@github.com:
…
Thanks for your reply. However, there are only 50 scenes in category annotation (http://103.24.77.34:8080/scenenn/contrib/nyu_class/) while CVPR'18 preprocessed data provides 76 scenes. Is there 76 scenes category annotation (unpreprocessed) dataset link? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#9?email_source=notifications&email_token=AABBPMVRTO4VGND472P5B5DQHYDFXA5CNFSM4IS6MMO2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5XG2PI#issuecomment-527330621>, or mute the thread https://github.com/notifications/unsubscribe-auth/AABBPMRRQYEKRPEQ6QAV4U3QHYDFXANCNFSM4IS6MMOQ .

Here is the temporary link for 76 scenes in ply format. This version has all data channels (color, instance, category annotation) all in a single ply.
https://drive.google.com/file/d/1d_ILfaxpJBpiiwCZtvC4jEKnixEr9N2l/view

Please check.