gomatch
gomatch copied to clipboard
BPnPNet Results
Hi,
I noticed that in the GoMatch paper, the BPnPNet results are quite a bit worse than the numbers reported in their paper. Looking at the following two tables:
BPnPNet reported results for MegaDepth test
GoMatch reported results for MegaDepth test
I guess the retrieval setup might be somewhat different and you retrained the model (maybe due to a different split?), but the difference seems quite large (>10x worse).
Any thoughts what caused this discrepancy?
Thank you 🙂