1. OSGDecoder
Triplane decoder that gives RGB and sigma values from sampled features.
Concatenate triplane features(3 * n_feature) to hidden_dim
linear layers(hd)
output → (hd → 1 + 3)
batch, n_planes, _M, _C = sampled_features.shape
[batch_size * num_samples, n_planes * n_features] → net() →
[batch_size, num_samples(M), output_dim]
sigmoid to clamp RGB
return RGB, sigma
2. Triplane Synthesizer
Synthesizer that renders a triplane volume with planes and a camera.
# planes: (N, 3, D', H', W')
# cameras: (N, M, D_cam) # d_cam 包含25维 4 * 4 + 3 * 3(intrinsics)
# anchors: (N, M, 2)
# resolutions: (N, M, 1)
# bg_colors: (N, M, 1)
# region_size: int- ray_sample
Create rays based on camera location and intrinsic
(cam2world, intrinsics, resolution, anchor(-1, 2))
→ (ray_origins(N * M, origin_loc_dim), ray_directions)
- volume rendering
Kiss your perfect day goodbye —>