mirror of
https://github.com/nerfstudio-project/nerfstudio.git
synced 2026-11-11 18:44:21 +08:00
7e9a56fac4
* starting docs for pipelines * working on an example * clearing output * starting to implement config changes and add documentation * wip: removed cicular imports; porting basic config into python; todo: get python setups workin * circular imports again * wip: pausing at dataloader sort of working * break for today * running vanilla * wip need to get all the config docs working * laptop about to die * instant ngp working * added all configs for all methods; not tested yet: [mipnerf360, nerfw, semantic_nerf] * unpacking later * done with pylinting * mostly done. just need to do docs and one test * added import order to docs * fixing near/far plane and other config things * making loggers also instantiable via configs * docs * lint * making --help work with dcargs * moving configs back into flat * convert list to tuples; fixed post init function with mutable params * lint * made run_eval and benchmarking scripts compatible with new config/dcargs * lint * added option to load configs from presaved config Co-authored-by: evonneng <evonne_ng@berkeley.edu>
254 lines
8.7 KiB
Python
254 lines
8.7 KiB
Python
# Copyright 2022 The Plenoptix Team. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
"""
|
|
Some ray datastructures.
|
|
"""
|
|
import random
|
|
from dataclasses import dataclass
|
|
from typing import Optional
|
|
|
|
import torch
|
|
from torchtyping import TensorType
|
|
|
|
from nerfactory.utils.math import Gaussians, conical_frustum_to_gaussian
|
|
from nerfactory.utils.tensor_dataclass import TensorDataclass
|
|
|
|
|
|
@dataclass
|
|
class Frustums(TensorDataclass):
|
|
"""Describes region of space as a frustum.
|
|
|
|
Attributes:
|
|
origins: xyz coordinate for ray origin.
|
|
directions: Direction of ray.
|
|
starts: Where the frustum starts along a ray.
|
|
ends: Where the frustum ends along a ray.
|
|
pixel_area: Projected area of pixel a distance 1 away from origin.
|
|
"""
|
|
|
|
origins: TensorType["bs":..., 3]
|
|
directions: TensorType["bs":..., 3]
|
|
starts: TensorType["bs":..., 1]
|
|
ends: TensorType["bs":..., 1]
|
|
pixel_area: TensorType["bs":..., 1]
|
|
|
|
def get_positions(self) -> TensorType[..., 3]:
|
|
"""Calulates "center" position of frustum. Not weighted by mass.
|
|
|
|
Returns:
|
|
TensorType[..., 3]: xyz positions.
|
|
"""
|
|
return self.origins + self.directions * (self.starts + self.ends) / 2
|
|
|
|
def get_gaussian_blob(self) -> Gaussians:
|
|
"""Calculates of guassian approximation of conical frustum.
|
|
|
|
Resturns:
|
|
Gaussians: Conical frustums approximated by gaussian distribution.
|
|
"""
|
|
# Cone radius is set such that the square pixel_area matches the cone area.
|
|
cone_radius = torch.sqrt(self.pixel_area) / 1.7724538509055159 # r = sqrt(pixel_area / pi)
|
|
return conical_frustum_to_gaussian(
|
|
origins=self.origins,
|
|
directions=self.directions,
|
|
starts=self.starts,
|
|
ends=self.ends,
|
|
radius=cone_radius,
|
|
)
|
|
|
|
@classmethod
|
|
def get_mock_frustum(cls, device="cpu") -> "Frustums":
|
|
"""Helper function to generate a placeholder frustum.
|
|
|
|
Returns:
|
|
Frustums: A size 1 frustum with meaningless values.
|
|
"""
|
|
return Frustums(
|
|
origins=torch.ones((1, 3)).to(device),
|
|
directions=torch.ones((1, 3)).to(device),
|
|
starts=torch.ones((1, 1)).to(device),
|
|
ends=torch.ones((1, 1)).to(device),
|
|
pixel_area=torch.ones((1, 1)).to(device),
|
|
)
|
|
|
|
|
|
@dataclass
|
|
class RaySamples(TensorDataclass):
|
|
"""Samples along a ray
|
|
|
|
Args:
|
|
frustums: Frustums along ray.
|
|
camera_indices: Camera index.
|
|
valid_mask: Rays that are valid.
|
|
deltas: "width" of each sample.
|
|
ts: sample-camera distance of each sample.
|
|
"""
|
|
|
|
frustums: Frustums
|
|
camera_indices: Optional[TensorType["bs":..., 1]] = None
|
|
valid_mask: Optional[TensorType["bs":..., 1]] = None
|
|
deltas: Optional[TensorType["bs":..., 1]] = None
|
|
|
|
def get_weights(self, densities: TensorType[..., "num_samples", 1]) -> TensorType[..., "num_samples", 1]:
|
|
"""Return weights based on predicted densities
|
|
|
|
Args:
|
|
densities (TensorType[..., "num_samples", 1]): Predicted densities for samples along ray
|
|
|
|
Returns:
|
|
TensorType[..., "num_samples", 1]: Weights for each sample
|
|
"""
|
|
|
|
delta_density = self.deltas * densities
|
|
alphas = 1 - torch.exp(-delta_density)
|
|
|
|
# TODO: put both versions of the transmittance calculation somewhere
|
|
# mip-nerf version of transmittance calculation:
|
|
transmittance = torch.cumsum(delta_density[..., :-1, :], dim=-2)
|
|
transmittance = torch.cat(
|
|
[torch.zeros((*transmittance.shape[:1], 1, 1)).to(densities.device), transmittance], dim=-2
|
|
)
|
|
transmittance = torch.exp(-transmittance) # [..., "num_samples"]
|
|
|
|
# most nerf codebases do the following:
|
|
# transmittance = torch.cat(
|
|
# [torch.ones((*alphas.shape[:1], 1)).to(densities.device), 1.0 - alphas + 1e-10], dim=-1
|
|
# )
|
|
# transmittance = torch.cumprod(transmittance, dim=-1)[..., :-1] # [..., "num_samples"]
|
|
|
|
weights = alphas * transmittance # [..., "num_samples"]
|
|
|
|
return weights
|
|
|
|
def set_valid_mask(self, valid_mask: TensorType[..., "num_samples"]) -> None:
|
|
"""Sets valid mask"""
|
|
self.valid_mask = valid_mask
|
|
|
|
def apply_masks(self) -> "RaySamples":
|
|
"""Use valid_mask to mask samples.
|
|
|
|
Returns:
|
|
RaySamples: New set of masked samples.
|
|
"""
|
|
if self.valid_mask is not None:
|
|
return self[self.valid_mask[..., 0]]
|
|
return self
|
|
|
|
|
|
@dataclass
|
|
class RayBundle(TensorDataclass):
|
|
"""A bundle of ray parameters.
|
|
|
|
Args:
|
|
origins: Ray origins (XYZ)
|
|
directions: Unit ray direction vector
|
|
pixel_area: Projected area of pixel a distance 1 away from origin
|
|
camera_indices: Camera indices
|
|
nears: Distance along ray to start sampling
|
|
fars: Rays Distance along ray to stop sampling
|
|
valid_mask: Rays that are valid
|
|
num_rays_per_chunk: Number of rays per chunk
|
|
"""
|
|
|
|
# TODO(ethan): make sure the sizes with ... are correct
|
|
origins: TensorType[..., 3]
|
|
directions: TensorType[..., 3]
|
|
pixel_area: TensorType[..., 1]
|
|
camera_indices: Optional[TensorType[..., 1]] = None
|
|
nears: Optional[TensorType[..., 1]] = None
|
|
fars: Optional[TensorType[..., 1]] = None
|
|
valid_mask: Optional[TensorType[..., 1, bool]] = None
|
|
num_rays_per_chunk: int = 128
|
|
|
|
def set_camera_indices(self, camera_index: int) -> None:
|
|
"""Sets all of the the camera indices to a specific camera index.
|
|
|
|
Args:
|
|
camera_index (int): Camera index.
|
|
"""
|
|
self.camera_indices = torch.ones_like(self.origins[..., 0:1]).long() * camera_index
|
|
|
|
def __len__(self):
|
|
num_rays = torch.numel(self.origins) // self.origins.shape[-1]
|
|
return num_rays
|
|
|
|
def sample(self, num_rays: int) -> "RayBundle":
|
|
"""Returns a RayBundle as a subset of rays.
|
|
|
|
Args:
|
|
num_rays (int): Number of rays in output RayBundle
|
|
|
|
Returns:
|
|
RayBundle: RayBundle with subset of rays.
|
|
"""
|
|
assert num_rays <= len(self)
|
|
indices = random.sample(range(len(self)), k=num_rays)
|
|
return self[indices]
|
|
|
|
def get_row_major_sliced_ray_bundle(self, start_idx: int, end_idx: int) -> "RayBundle":
|
|
"""Flattens RayBundle and extracts chunk given start and end indicies.
|
|
|
|
Args:
|
|
start_idx (int): Start index of RayBundle chunk.
|
|
end_idx (int): End index of RayBundle chunk.
|
|
|
|
Returns:
|
|
RayBundle: Flattened RayBundle with end_idx-start_idx rays.
|
|
|
|
"""
|
|
return self.flatten()[start_idx:end_idx]
|
|
|
|
def get_ray_samples(
|
|
self, bin_starts: TensorType["bs":..., "num_samples", 1], bin_ends: TensorType["bs":..., "num_samples", 1]
|
|
) -> RaySamples:
|
|
"""Produces samples for each ray by projection points along the ray direction.
|
|
|
|
|
|
Args:
|
|
bin_starts (TensorType[..., "num_samples", 1]): Distance from origin to start of bin.
|
|
bin_ends (TensorType[..., "num_samples", 1]): Distance from origin to end of bin.
|
|
|
|
Returns:
|
|
RaySamples: Samples projected along ray.
|
|
"""
|
|
device = self.origins.device
|
|
|
|
valid_mask = torch.ones((bin_starts.shape), dtype=torch.bool, device=device)
|
|
|
|
dists = bin_ends - bin_starts # [..., N_samples, 1]
|
|
deltas = dists * torch.norm(self.directions[:, :], dim=-1)[..., None, None]
|
|
|
|
if self.camera_indices is not None:
|
|
camera_indices = self.camera_indices[..., None]
|
|
else:
|
|
camera_indices = None
|
|
|
|
frustums = Frustums(
|
|
origins=self.origins[..., None, :], # [..., 1, 3]
|
|
directions=self.directions[..., None, :], # [..., 1, 3]
|
|
starts=bin_starts, # [..., N_samples, 1]
|
|
ends=bin_ends, # [..., N_samples, 1]
|
|
pixel_area=self.pixel_area[..., None, :], # [..., 1, 1]
|
|
).to(device)
|
|
|
|
ray_samples = RaySamples(
|
|
frustums=frustums,
|
|
camera_indices=camera_indices, # [..., 1, 1]
|
|
valid_mask=valid_mask, # [..., N_samples, 1]
|
|
deltas=deltas, # [..., N_samples, 1]
|
|
)
|
|
|
|
return ray_samples
|