search

bailubin / MMGR

Official implementation of the paper "Geographic mapping with unsupervised multi-modal representation learning from VHR images and POIs"
18 stars 1 forks source link

Question about dataset generation as mmgrdataset #3

Open BioDPJ opened 5 months ago

BioDPJ commented 5 months ago

Dr. Bai, Good evening. I've debugged the MMGR codes based on the released mmgrdataset. Now I want to train and test on dataset in another city. According to your reference, I visited the required 'POI_ZoneID.csv' and 'edge_index_weight.csv' in repository 'SemanticPOIEmbedding'. Here's my question: How to generate 'edge_index_weight.csv' for DT network construction? I searched the related papers from you and Dr. Huang while got nothing operatable. I would indeed appreciate a lot if hearing from you as soon as possible. Looking forward to your response and really thank you for your help.

Best regards, BioDPJ

BioDPJ commented 5 months ago

If it doesn't bother you, could we contact via email?

bailubin commented 5 months ago

Thank you for your interest. The 'edge_index_weight.csv' is gegnerated bt the construction of Delaunay Network. It recordes the edge weight of the Delaunay Network, you can refer to the paper " Estimating urban functional distributions with semantics preserved POI embedding". In MMGR paper, we only use the inverse-distance transition bias described in this paper. image

As for the code inmpetation, you can use Delaunay function from scipy.spatial, and then calculate the edge weight (transition bias).

BioDPJ commented 5 months ago

Got it, Dr. Bai. Thank you for your help! I'll proceed with it and remain the solution available for future discussion. Looking forward to staying in touch :)