Yawar Nihal Siddiqui
Yawar Nihal Siddiqui
Hi, its very probably that at some point I switched from using npys to npzs because of space issues and didnt document / update the code (can't exactly remember since...
Hi, on the server where I was training this, I had access to large amounts of ram (>240G). There's a flag that prevents loading into the memory (preload_scenes and preload_retrievals)....
Hi, I dont have a checkpoint anymore. Can you elaborate on the issues you're facing now?
Hi, There is a slight performance drop during the retrieval attention phase (phase 2) where we freeze the rest of the networks, but should not be too much. On shapenet...
> 1. Performance 1. nearest neighbour: iou:=0,581, cd:=1.104, precision:=0.714, recall:=0.733, One thing that strikes odd is that the retrievals here seem quite bad; is this from the checked in retrievals...
Theoretically should be the same (network or the ones checked in), but just to be 100% sure i would try the one's checked in (maybe its quicker to just evaluate...
Glad it helps! I was going through my early logs for this experiment, and I found this one table in the notes on some early scene level experiments.  The...
Hi, The ones during training are just a rough estimate, since the evaluation is too expensive to do during the training. Regards, Yawar Nihal On Sun 19. May 2024 at...
Thank you! Will fix it :)
Hi guys, we're still waiting for a legal approval from our collaborators for releasing the code and the models.