A few questions regarding your paper about convolutional_occupancy_networks HOT 4 CLOSED

@Na-Z Thanks for your interest!

Hi, thanks for sharing your great work!

I have a few questions:

1. You mentioned that "our method is independent of the input representation" in the paper. I am curious to know that if it is possible to use multi-view images as input in your framework. If the answer is yes, could you provide a hint how to implement that?

We have not tackled the multi-view image inputs yet, but you can check out some recent papers like this and this, where they show how to fuse multi-view images to a feature volume.

2. If train your method in a real-world dataset and test on another real-world dataset, do you think it will still work?

Sure, if you manage to generate such a real-world dataset for training, I think it should work and work better for real-world scenarios ;)

3. As mentioned in Sec. 6 of the supplementary material, you "randomly sample one point within the scene as the center of the crop". May I know how you balance the positive and negative samples (i.e., gt occupancy value 1 and 0) with random sampling?

I did not check the balance between positive and negative samples, but I think we definitely have more negatives (free space).

Looking forward to your reply. Thanks.

Hope I answer your questions.

Best,
Songyou

from convolutional_occupancy_networks.

Hi Songyou,

Thanks for your prompt reply.

May I know why not use ScanNet for training in your experiments? Any concern?

Best,
Na

from convolutional_occupancy_networks.

@Na-Z Because the ScanNet data is noisy and incomplete, plus we found that it generalizes already pretty well using the synthetic dataset training.

from convolutional_occupancy_networks.

I see. Thanks.

from convolutional_occupancy_networks.