dataset_inputs -> dataparser_outputs (#627)

This commit is contained in:
Matthew Tancik authored and GitHub committed 2022-10-03 09:45:36 -07:00
1 parent ef5384afb3
commit 3efd3c3298
17 files changed
+52 -179

No files matched your search

-79
View File
@@ -1,79 +0,0 @@
# Dataset configuration
#### Specifying dataset
The dataset path and type is specified in the model config. Both `dataset_inputs_train` and a `dataset_inputs_eval` need to be set. You can see `configs/graph_default.yaml` for more details on what a full config would look like.
```yaml
data:
dataset_inputs_train:
data_directory: data/instant_ngp/fox
dataset_format: instant_ngp
alpha_color: white
dataset_inputs_eval:
data_directory: data/instant_ngp/fox
dataset_format: instant_ngp
alpha_color: white
```
#### Visualize dataset
TODO
#### Supported datasets
We include support for a set of standard dataset formats.
###### Blender Synthetic
```{button-link} https://drive.google.com/drive/u/1/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1
:color: primary
:outline:
Download
```
```yaml
dataset_format: blender
```
Classic NeRF dataset including the "lego bulldozer" scene that was released with the original NeRF paper.
###### Instant NGP
```{button-link} https://github.com/NVlabs/instant-ngp#nerf-fox
:color: primary
:outline:
Download
```
```yaml
dataset_format: instant_ngp
```
###### MipNeRF 360
```{button-link} https://jonbarron.info/mipnerf360/
:color: primary
:outline:
Download
```
```yaml
dataset_format: mipnerf_360
```
###### Record3D
```{button-link} https://record3d.app/
:color: primary
:outline:
Download
```
```yaml
dataset_format: record_3d
```
Directly import dataset recorded from a >= iPhone 12 Pro using the Record3D app.
Record a video and export with the `EXR + JPG sequence` format. Unzip export and `rgb` folder before training.
@@ -1,46 +0,0 @@
# Custom dataset formats
We've implemented many common datasets inside the `nerfstudio/format/<dataset_format>.py` files. If you have a posed image dataset that does not match these existing formats can create a custom dataset format.
1. Create a function `nerfstudio/format/<dataset_format>.py` and implements a function called `load_<dataset_format>_data` returns a `DatasetInputs` object.
```python
# The DatasetInputs dataclass that is passed around to create datasets and initialize the graphs.
# See `nerfstudio/data/structs.py` for the code.
@dataclass
class DatasetInputs:
"""Dataset inputs are used to initialize datasets and the NeRF graph."""
image_filenames: List[str]
intrinsics: torch.tensor = None
camera_to_world: torch.tensor = None
mask_filenames: List[str] = None
depth_filenames: List[str] = None
scene_box: SceneBox = SceneBox()
semantics: Semantics = Semantics()
point_cloud: PointCloud = PointCloud()
alpha_color: Optional[TensorType[3]] = None
# The method signature for the `instant_ngp` dataset format.
# See `nerfstudio/data/format/instant_ngp.py` for the code.
def load_instant_ngp_data(
basedir: str,
split: str = "train", camera_translation_scalar=0.33
) -> DatasetInputs:
"""Returns a DatasetInputs struct."""
```
2. Add a new case to `get_dataset_inputs()` in `nerfstudio/data/utils.py` for the new dataset.
```python
# The method signature for get_dataset_inputs.
# Notice that the parameters match the yaml config parameters above.
# See `nerfstudio/data/utils.py` for the code.
def get_dataset_inputs(
data_directory: str,
dataset_format: str,
split: str,
alpha_color: Optional[Union[str, list, ListConfig]] = None,
) -> DatasetInputs:
"""Makes a call to `load_<dataset_format>_data` and returns a DatasetInputs struct."""
```
-2
View File
@@ -4,6 +4,4 @@ Data Setup
.. toctree::
:maxdepth: 1
choosing_dataset.md
creating_dataset.md
custom_dataset_format.md