nerfstudio/nerfactory/models/instant_ngp.py

# 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.

"""
Implementation of Instant NGP.
"""

from __future__ import annotations

from dataclasses import dataclass, field
from typing import Dict, List, Literal, Optional, Tuple, Type

import nerfacc
import torch
from torch.nn import Parameter
from torchmetrics import PeakSignalNoiseRatio
from torchmetrics.functional import structural_similarity_index_measure
from torchmetrics.image.lpip import LearnedPerceptualImagePatchSimilarity

from nerfactory.cameras.rays import RayBundle
from nerfactory.fields.instant_ngp_field import field_implementation_to_class
from nerfactory.fields.modules.field_heads import FieldHeadNames
from nerfactory.models.base import Model, ModelConfig
from nerfactory.models.modules.ray_sampler import VolumetricSampler
from nerfactory.optimizers.loss import MSELoss
from nerfactory.renderers.renderers import (
    AccumulationRenderer,
    DepthRenderer,
    RGBRenderer,
)
from nerfactory.utils import colors, visualization
from nerfactory.utils.callbacks import (
    TrainingCallback,
    TrainingCallbackAttributes,
    TrainingCallbackLocation,
)


@dataclass
class InstantNGPModelConfig(ModelConfig):
    """Instant NGP Model Config"""

    _target: Type = field(
        default_factory=lambda: NGPModel
    )  # We can't write `NGPModel` directly, because `NGPModel` doesn't exist yet
    """target class to instantiate"""
    enable_collider: bool = False
    """Whether to create a scene collider to filter rays."""
    collider_params: Optional[Dict[str, float]] = None
    """Instant NGP doesn't use a collider."""
    field_implementation: Literal["torch", "tcnn"] = "tcnn"  # torch, tcnn, ...
    """one of "torch" or "tcnn", or other fields in 'field_implementation_to_class'"""
    max_num_samples_per_ray: int = 1024
    """Number of samples in field evaluation."""
    grid_resolution: int = 128
    """Resolution of the grid used for the field."""
    cone_angle: float = 0.0
    """Should be set to 0.0 for blender scenes but 1./256 for real scenes."""
    near_plane: float = 0.05
    """How far along ray to start sampling."""
    randomize_background: bool = False
    """Whether to randomize the background color."""


class NGPModel(Model):
    """Instant NGP model

    Args:
        config: instant NGP configuration to instantiate model
    """

    config: InstantNGPModelConfig

    def __init__(self, config: InstantNGPModelConfig, **kwargs) -> None:
        assert config.field_implementation in field_implementation_to_class
        self.field = None
        super().__init__(config=config, **kwargs)

    def populate_modules(self):
        """Set the fields and modules."""
        super().populate_modules()

        # torch or tiny-cuda-nn version
        self.field = field_implementation_to_class[self.config.field_implementation](self.scene_bounds.aabb)

        self.scene_aabb = Parameter(self.scene_bounds.aabb.flatten(), requires_grad=False)

        # Sampler
        self.sampler = VolumetricSampler(
            aabb=self.scene_aabb,
            density_fn=self.field.density_fn,
            grid_resolution=self.config.grid_resolution,
            num_samples=self.config.max_num_samples_per_ray,
        )

        # renderers
        background_color = None if self.config.randomize_background else colors.WHITE
        self.renderer_rgb = RGBRenderer(background_color=background_color)
        self.renderer_accumulation = AccumulationRenderer()
        self.renderer_depth = DepthRenderer()

        # losses
        self.rgb_loss = MSELoss()

        # metrics
        self.psnr = PeakSignalNoiseRatio(data_range=1.0)
        self.ssim = structural_similarity_index_measure
        self.lpips = LearnedPerceptualImagePatchSimilarity()

    def get_training_callbacks(
        self, training_callback_attributes: TrainingCallbackAttributes
    ) -> List[TrainingCallback]:
        return [
            TrainingCallback(
                where_to_run=[TrainingCallbackLocation.BEFORE_TRAIN_ITERATION],
                update_every_num_iters=1,
                func=self.sampler.occ_field.every_n_step,
            ),
        ]

    def get_param_groups(self) -> Dict[str, List[Parameter]]:
        param_groups = {}
        if self.field is None:
            raise ValueError("populate_fields() must be called before get_param_groups")
        param_groups["fields"] = list(self.field.parameters())
        return param_groups

    def get_outputs(self, ray_bundle: RayBundle):
        assert self.field is not None
        num_rays = len(ray_bundle)

        with torch.no_grad():
            ray_samples, packed_info, ray_indices = self.sampler(
                ray_bundle=ray_bundle,
                near_plane=self.config.near_plane,
            )

        field_outputs = self.field(ray_samples)

        # accumulation
        weights = nerfacc.volumetric_rendering_weights(
            packed_info=packed_info,
            sigmas=field_outputs[FieldHeadNames.DENSITY],
            frustum_starts=ray_samples.frustums.starts,
            frustum_ends=ray_samples.frustums.ends,
        )

        rgb = self.renderer_rgb(
            rgb=field_outputs[FieldHeadNames.RGB],
            weights=weights,
            ray_indices=ray_indices,
            num_rays=num_rays,
        )
        depth = self.renderer_depth(
            weights=weights, ray_samples=ray_samples, ray_indices=ray_indices, num_rays=num_rays
        )
        accumulation = self.renderer_accumulation(weights=weights, ray_indices=ray_indices, num_rays=num_rays)
        alive_ray_mask = accumulation.squeeze(-1) > 0

        outputs = {
            "rgb": rgb,
            "accumulation": accumulation,
            "depth": depth,
            "alive_ray_mask": alive_ray_mask,  # the rays we kept from sampler
            "num_samples_per_ray": packed_info[:, 1],
        }
        return outputs

    def get_metrics_dict(self, outputs, batch):
        image = batch["image"].to(self.device)
        metrics_dict = {}
        metrics_dict["psnr"] = self.psnr(outputs["rgb"], image)
        metrics_dict["num_samples_per_batch"] = outputs["num_samples_per_ray"].sum()
        return metrics_dict

    def get_loss_dict(self, outputs, batch, metrics_dict=None):
        image = batch["image"].to(self.device)
        mask = outputs["alive_ray_mask"]
        rgb_loss = self.rgb_loss(image[mask], outputs["rgb"][mask])
        loss_dict = {"rgb_loss": rgb_loss}
        return loss_dict

    def get_image_metrics_and_images(
        self, outputs: Dict[str, torch.Tensor], batch: Dict[str, torch.Tensor]
    ) -> Tuple[Dict[str, float], Dict[str, torch.Tensor]]:

        image = batch["image"].to(self.device)
        rgb = outputs["rgb"]
        acc = visualization.apply_colormap(outputs["accumulation"])
        depth = visualization.apply_depth_colormap(
            outputs["depth"],
            accumulation=outputs["accumulation"],
        )
        alive_ray_mask = visualization.apply_colormap(outputs["alive_ray_mask"])

        combined_rgb = torch.cat([image, rgb], dim=1)
        combined_acc = torch.cat([acc], dim=1)
        combined_depth = torch.cat([depth], dim=1)
        combined_alive_ray_mask = torch.cat([alive_ray_mask], dim=1)

        # Switch images from [H, W, C] to [1, C, H, W] for metrics computations
        image = torch.moveaxis(image, -1, 0)[None, ...]
        rgb = torch.moveaxis(rgb, -1, 0)[None, ...]

        psnr = self.psnr(image, rgb)
        ssim = self.ssim(image, rgb)
        lpips = self.lpips(image, rgb)

        # all of these metrics will be logged as scalars
        metrics_dict = {"psnr": float(psnr.item()), "ssim": float(ssim), "lpips": float(lpips)}  # type: ignore
        # TODO(ethan): return an image dictionary

        images_dict = {
            "img": combined_rgb,
            "accumulation": combined_acc,
            "depth": combined_depth,
            "alive_ray_mask": combined_alive_ray_mask,
        }

        return metrics_dict, images_dict