search

chanyn / HKRM

Hybrid Knowledge Routed Module for Large-scale Object Detection (NerIPS2018)
103 stars 19 forks source link

About datasets #6

Open Easyfeng222 opened 5 years ago

Easyfeng222 commented 5 years ago

Excuse me ,How can I make vg_attr_frequency_1000.pkl and vg_pair_frequency_1000.pkl?

Easyfeng222 commented 5 years ago

I am very interested in how you get vg_attr_frequency_1000.pkl. Can you tell me how to get it from the VG dataset? thank you very much! ! @chanyn

chanyn commented 5 years ago

In brief, we sum each object attribute annotation of entire dataset if the class of object is in the top 1000 frequent categories we had defined, so we get a 1000D frequency statistic matrix. For relationship frequency statistic matrix, we first new a 1000x1000 zero matrix Qr. Then we check the ground-truth of each relationship annotation and add 1 to the corresponding position of Qr, if both classes are in the range of top 1000 frequent categories. Qr is vg_pair_frequency_1000.pkl that we have provided. For attribute frequency statistic matrix, we new a 1000x200 zero matrix Qa. We just consider top 200 frequent attribute in VG. Then we check the gt annotation of each object and add 1 to the corresponding position of Qa. It's similar to build Qr. Finally, Qa is vg_attr_frequency_1000.pkl.

Easyfeng222 commented 5 years ago

@chanyn I am sorry to see and express my thanks so late. I still have a problem, I want to rebuild a database training, how can I get gt, coco gt is invalid.