search

rowanz / neural-motifs

Code for Neural Motifs: Scene Graph Parsing with Global Context (CVPR 2018)
https://rowanzellers.com/neuralmotifs
MIT License
525 stars 115 forks source link

how to re-implement the result in no graph constraints ??? #19

Closed akira-l closed 6 years ago

akira-l commented 6 years ago

I got the result in graph constraints with given code and parameters, but where is the no constraints part? How to implement no graph constraints ?

Check relation index (rel_inds in rel_model.py) and make sure every proposal only have one relation to another proposal area ?

rowanz commented 6 years ago

Use the -multipred option - let me know if that doesn’t work :)

On Tue, Aug 14, 2018 at 7:11 AM SaiYiKi notifications@github.com wrote:

I got the result in graph constraints with given code and parameters, but where is the no constraints part? How to implement no graph constraints ?

Check relation index (rel_inds in rel_model.py) and make sure every proposal only have one relation to another proposal area ?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/rowanz/neural-motifs/issues/19, or mute the thread https://github.com/notifications/unsubscribe-auth/ABWJxx0zJlBWfHBpYN-l70zhIwhqhEHaks5uQtqPgaJpZM4V8g6O .

akira-l commented 6 years ago

OK THANKS!

akira-l commented 6 years ago

oh, by the way, those checkpoints given in your google drives were trained in graph constraints ? Should I train again with no graph constraints ?

akira-l commented 6 years ago

oh, my mistake, may be the with/without graph constraints just like reranking in retrieval. The constraints are not relevent to the network model training, right?