search

yashbhalgat / Contrastive-Lift

[NeurIPS 2023 Spotlight] Code for "Contrastive Lift: 3D Object Instance Segmentation by Slow-Fast Contrastive Fusion"
https://www.robots.ox.ac.uk/~vgg/research/contrastive-lift/
59 stars 1 forks source link

When will the code be available? #1

Closed Runsong123 closed 11 months ago

Runsong123 commented 11 months ago

Dear author,

Thanks for your great work. I want to test your methods in my experiments and I want to know when the code will be available? Besides, Does the proposed method take the similar time as panoptic-lifting for optimization time on ScanNet. BTW, I found the panoptic-lifting takes 36 hours on my single 3090 RTX for one ScanNet scene (042302). Looking forward to your reply. Many thanks in advance.

Bests, Runsong

yashbhalgat commented 11 months ago

Hi Runsong, thank you for your interest in my work! I will be releasing the code over the weekend along with some pretrained checkpoints. Stay tuned!

The loss functions used by Contrastive-Lift are the same as Panoptic-Lifting, except of course the instance branch loss. And the most time-consuming loss is the "segment consistency" loss, which is present in both methods (ours and PanopLi). I found that you don't need train the model for as many epochs as PanopLi, so the training can be done in around 20-24 hours for Contrastive-Lift. Hope this answers your question!

Runsong123 commented 11 months ago

I see. Thank you very much for your explanation !!