Network details

[yangyuankun865]

Hello. Thanks for your great work and your release of code. I am new to research and am understanding your model. In the paper, you mentioned 'neural feature is a 59-channel vector' (8+16+32 multi-scale +3). However, when I checked details of neural features in point_aggregators.py on Nerf dataset, the input has only 32 channels, so finally your project did not apply multi-scale? Besides, your model applied block2 and block3 which integrated features with colors and directions. Is it another modification after publishing the paper? It's better to clarify modifications after publishing the paper to avoid misunderstandings.

[Xharlie]

hi, we reduce the channel numbers for better latency