No project description provided
Project description
GeoSynth: A Photorealistic Synthetic Indoor Dataset for Scene Understanding
Deep learning has revolutionized many scene perception tasks over the past decade. Some of these improvements can be attributed to the development of large labeled datasets. The creation of such datasets can be an expensive, time-consuming, and is an imperfect process. To address these issues, we introduce GeoSynth, a diverse photorealistic synthetic dataset for indoor scene perception tasks. Each GeoSynth exemplar contains rich labels, including segmentation, geometry, camera parameters, surface material, lighting, and more. We demonstrate that supplementing real training data with GeoSynth can significantly improve network performance on perception tasks, like semantic segmentation.
Watch our presentation from IEEE VR 2023.
GeoSynth is used internally at Geomagical Labs to help power the IKEA Kreativ home design experience.
Installation
GeoSynth requires Python >=3.8
and can be installed via:
pip install geosynth
This installs:
- The
geosynth
python library, providing a pythonic interface for reading and processing GeoSynth data. - The
geosynth
command-line tool, which offers a convenient way of downloading the geosynth dataset.
Some optional visualization tools will require matplotlib
.
Dataset Download
Attention! Currently only the demo
variant is available, the full dataset and additional datatypes will be released in the near future.
To download just a few scenes of the dataset, download the demo
variant.
The demo
variant is the default --variant
option:
geosynth download non-hdr --variant=demo
If you also wish to include HDR data, specify all
, instead.
The HDR data more than doubles the size of the download, so only download it if you need it.
It is recommended to only specify the data types you need.
By default, the contents will be downloaded to ~/data/geosynth/
.
To specify an alternative download location, specify the --dst
argument.
See all download options by running geosynth download --help
:
$ geosynth download --help
Usage: geosynth download [OPTIONS] DTYPES...
Download the GeoSynth data.
╭─ Arguments ──────────────────────────────────────────────────────────────────────────╮
│ * dtypes DTYPES... Assets to download. Either specify "non-hdr", "all", or │
│ a subset of: [cube_environment_map, depth, extrinsics, │
│ gravity, hdr_cube_environment_map, hdr_reflectance, │
│ hdr_residual, hdr_rgb, hdr_shading, │
│ hdr_sphere_environment_map, instance_segmentation, │
│ intrinsics, layout_lines_full, layout_lines_occluded, │
│ layout_lines_visible, lighting, normals, reflectance, │
│ residual, rgb, semantic_segmentation, shading, │
│ sphere_environment_map]). │
│ [required] │
╰──────────────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────────────╮
│ --dst PATH GeoSynth download directory. │
│ [default: ~/data/geosynth] │
│ --variant [demo|full] Variant of dataset to download. │
│ [default: demo] │
│ --force --no-force Force a re-download, despite locally │
│ cached files. │
│ [default: no-force] │
│ --cleanup --no-cleanup Delete zip files after unzipping. │
│ [default: cleanup] │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────────────╯
Usage
Once the dataset has been downloaded, data can be accessed in python:
from geosynth import GeoSynth
geosynth = GeoSynth("PATH_TO_DATA") # or leave empty for default "~/data/geosynth/".
print(f"GeoSynth has {len(geosynth)} scenes.")
scene = geosynth[100] # Data can be accessed via indexing like a list.
# or iterated over in a for loop:
for scene in dataset:
# Each Scene object contains attributres for each datatype.
# Contents can be read from disk via the ``read`` method.
rgb = scene.rgb.read() # (H, W, 3) np.ndarray
depth = scene.depth.read() # (H, W) np.ndarray
intrinsics = scene.intrinsics.read() # (3, 3) camera intrinsics
instances = scene.instance_segmentation.read() # dictionary of instance masks.
# many datatypes have a ``visualize`` method
depth_viz = scene.depth.visualize(depth) # Returns a (H,W,3) turbo-colorized image.
instances_viz = scene.instance_segmentation.visualize(instances, rgb=rgb)
License
The GeoSynth code is released under the Apache-2.0 License.
The GeoSynth data provided at storage.googleapis.com/geomagical-geosynth-public
is available under the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) License.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file geosynth-1.0.1.tar.gz
.
File metadata
- Download URL: geosynth-1.0.1.tar.gz
- Upload date:
- Size: 26.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: poetry/1.7.1 CPython/3.11.4 Darwin/23.1.0
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | cde690e97ca569d80202a4d271a8bcbe84b12e693b57f874350d61f5acf36cd2 |
|
MD5 | a7b5b2b5e8a2721a65af5d3e5f3bb214 |
|
BLAKE2b-256 | 57d9ba95e2537743b03665c3700adcbbb93ce5dccc6f9279e032de3836493da3 |
File details
Details for the file geosynth-1.0.1-py3-none-any.whl
.
File metadata
- Download URL: geosynth-1.0.1-py3-none-any.whl
- Upload date:
- Size: 28.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: poetry/1.7.1 CPython/3.11.4 Darwin/23.1.0
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2e39fa154def5955c6ee3fa24704be2b07a552453eb47f21e8bc8072158392b3 |
|
MD5 | 97d8fc4017deb4c481bfca002b305564 |
|
BLAKE2b-256 | ff0020bd9b0cd7c9c807a0d938ce8f788c561e2b88c271887729c202c1fe3260 |