Question about the event number loss

[Tianbo-Pan]

Thanks for this thought-provoking work!

However, I have a question related to the event number loss.

pos = event_cur > 0
neg = event_cur < 0
loss_pos = torch.mean(((thres_pos * pos) - ((event_cur + 0.5) * pos)) ** 2)
loss_neg = torch.mean(((thres_neg * neg) - ((event_cur - 0.5) * neg)) ** 2)

why the event_cur needs to +/- 0.5?

[Tianbo-Pan]

In real-world datasets, how is the camera pose determined for novel view synthesis? I couldn’t locate the relevant code for selecting these viewpoints. Could you provide clarification on this? Thanks for your time and attention.