mirror of
https://github.com/nerfstudio-project/nerfstudio.git
synced 2026-11-11 02:24:38 +08:00
a8bce87ada
* pass grad sclaer from trainer to pipeline * lint * add grad scaler to arg doc --------- Co-authored-by: Matthew Tancik <tancik@berkeley.edu>
106 lines
4.5 KiB
Python
106 lines
4.5 KiB
Python
# Copyright 2022 the Regents of the University of California, Nerfstudio Team and contributors. 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.
|
|
|
|
"""
|
|
A pipeline that dynamically chooses the number of rays to sample.
|
|
"""
|
|
|
|
from dataclasses import dataclass, field
|
|
from typing import Literal, Type, Optional
|
|
|
|
import torch
|
|
from torch.cuda.amp.grad_scaler import GradScaler
|
|
|
|
from nerfstudio.data.datamanagers.base_datamanager import VanillaDataManager
|
|
from nerfstudio.pipelines.base_pipeline import VanillaPipeline, VanillaPipelineConfig
|
|
|
|
|
|
@dataclass
|
|
class DynamicBatchPipelineConfig(VanillaPipelineConfig):
|
|
"""Dynamic Batch Pipeline Config"""
|
|
|
|
_target: Type = field(default_factory=lambda: DynamicBatchPipeline)
|
|
target_num_samples: int = 262144 # 1 << 18
|
|
"""The target number of samples to use for an entire batch of rays."""
|
|
max_num_samples_per_ray: int = 1024 # 1 << 10
|
|
"""The maximum number of samples to be placed along a ray."""
|
|
|
|
|
|
class DynamicBatchPipeline(VanillaPipeline):
|
|
"""Pipeline with logic for changing the number of rays per batch."""
|
|
|
|
config: DynamicBatchPipelineConfig
|
|
datamanager: VanillaDataManager
|
|
dynamic_num_rays_per_batch: int
|
|
|
|
def __init__(
|
|
self,
|
|
config: DynamicBatchPipelineConfig,
|
|
device: str,
|
|
test_mode: Literal["test", "val", "inference"] = "val",
|
|
world_size: int = 1,
|
|
local_rank: int = 0,
|
|
grad_scaler: Optional[GradScaler] = None,
|
|
):
|
|
super().__init__(config, device, test_mode, world_size, local_rank)
|
|
assert isinstance(
|
|
self.datamanager, VanillaDataManager
|
|
), "DynamicBatchPipeline only works with VanillaDataManager."
|
|
|
|
self.dynamic_num_rays_per_batch = self.config.target_num_samples // self.config.max_num_samples_per_ray
|
|
self._update_pixel_samplers()
|
|
|
|
def _update_pixel_samplers(self):
|
|
"""Update the pixel samplers for train and eval with the dynamic number of rays per batch."""
|
|
if self.datamanager.train_pixel_sampler is not None:
|
|
self.datamanager.train_pixel_sampler.set_num_rays_per_batch(self.dynamic_num_rays_per_batch)
|
|
if self.datamanager.eval_pixel_sampler is not None:
|
|
self.datamanager.eval_pixel_sampler.set_num_rays_per_batch(self.dynamic_num_rays_per_batch)
|
|
|
|
def _update_dynamic_num_rays_per_batch(self, num_samples_per_batch: int):
|
|
"""Updates the dynamic number of rays per batch variable,
|
|
based on the total number of samples in the last batch of rays."""
|
|
self.dynamic_num_rays_per_batch = int(
|
|
self.dynamic_num_rays_per_batch * (self.config.target_num_samples / num_samples_per_batch)
|
|
)
|
|
|
|
def get_train_loss_dict(self, step: int):
|
|
model_outputs, loss_dict, metrics_dict = super().get_train_loss_dict(step)
|
|
|
|
# update the number of rays for the next step
|
|
if "num_samples_per_batch" not in metrics_dict:
|
|
raise ValueError(
|
|
"'num_samples_per_batch' is not in metrics_dict."
|
|
"Please return 'num_samples_per_batch' in the models get_metrics_dict function to use this method."
|
|
)
|
|
self._update_dynamic_num_rays_per_batch(int(metrics_dict["num_samples_per_batch"]))
|
|
self._update_pixel_samplers()
|
|
|
|
# add the number of rays
|
|
assert "num_rays_per_batch" not in metrics_dict
|
|
assert self.datamanager.train_pixel_sampler is not None
|
|
metrics_dict["num_rays_per_batch"] = torch.tensor(self.datamanager.train_pixel_sampler.num_rays_per_batch)
|
|
|
|
return model_outputs, loss_dict, metrics_dict
|
|
|
|
def get_eval_loss_dict(self, step: int):
|
|
model_outputs, loss_dict, metrics_dict = super().get_eval_loss_dict(step)
|
|
|
|
# add the number of rays
|
|
assert "num_rays_per_batch" not in metrics_dict
|
|
assert self.datamanager.eval_pixel_sampler is not None
|
|
metrics_dict["num_rays_per_batch"] = torch.tensor(self.datamanager.eval_pixel_sampler.num_rays_per_batch)
|
|
|
|
return model_outputs, loss_dict, metrics_dict
|