Update on Record3D Dataset (small fixes + documentation) (#286)

* updated dataset to deal with downscale factor and dataset size.

* added docs.

* updated resizing array to always guarantee size of dataset.
This commit is contained in:
Alexander Kristoffersen authored and GitHub committed 2022-08-22 17:15:18 -07:00
1 parent b471736a61
commit 07a0dbc3a1
2 files changed
+32 -4

No files matched your search

+15 -3
View File
@@ -386,12 +386,15 @@ class Record3D(Dataset):
downscale_factor: How much to downscale images. Defaults to 1.
val_skip: 1/val_skip images to use for validation. Defaults to 8.
aabb_scale: Scene scale, Defaults to 4.0.
max_dataset_size: Max number of images to train on. If the dataset has
more, images will be sampled approximately evenly. Defaults to 150.
"""
data_directory: str
downscale_factor: int = 1
val_skip: int = 8
aabb_scale = 4.0
max_dataset_size: int = 150
def _generate_dataset_inputs(self, split: str = "train") -> DatasetInputs:
abs_dir = get_absolute_path(self.data_directory)
@@ -406,6 +409,7 @@ class Record3D(Dataset):
for f in os.listdir(image_dir):
image_filenames.append(os.path.join(image_dir, f))
image_filenames = sorted(image_filenames, key=lambda fn: int(os.path.basename(fn)[: -len(ext)]))
image_filenames = np.array(image_filenames)
num_images = len(image_filenames)
metadata_path = os.path.join(abs_dir, "metadata.json")
@@ -418,6 +422,14 @@ class Record3D(Dataset):
axis=-1,
).astype(np.float32)
if num_images > self.max_dataset_size:
# Evenly select max_dataset_size images from dataset, including first
# and last indices.
idx = np.round(np.linspace(0, num_images - 1, self.max_dataset_size)).astype(int)
poses = poses[idx]
image_filenames = image_filenames[idx]
num_images = len(image_filenames)
# Normalization similar to Mipnerf360
poses = Mipnerf360.normalize_orientation(poses)
@@ -442,7 +454,7 @@ class Record3D(Dataset):
if num_images != poses.shape[0]:
raise RuntimeError(f"Different number of images ({num_images}), and poses ({poses.shape[0]})")
image_filenames = np.array(image_filenames)[idx]
image_filenames = image_filenames[idx]
poses = poses[idx]
# Centering poses
@@ -472,8 +484,8 @@ class Record3D(Dataset):
dataset_inputs = DatasetInputs(
image_filenames=image_filenames,
downscale_factor=1,
intrinsics=intrinsics,
downscale_factor=self.downscale_factor,
intrinsics=intrinsics * 1.0 / self.downscale_factor,
camera_to_world=camera_to_world,
scene_bounds=scene_bounds,
)