* undistortion should not occur if pinhole (smoothes the images), and accordingly fix rounding of downscaled image size
* rounding downscaling, more consistent with other resizing methods in public datasets
* reverting to cpu image caching
* re-allow camera unimplemented
* compatibility with mask/depth
* reverted scale factor rounding, corrected mask undistort condition
* moved distortion_params update inside PERSPECTIVE type if-clause
* bumping opencv to 4.8, to deal with undistortion bug with distortion parameters of zero
---------
Co-authored-by: liorf <liorf@amazon.com>
Co-authored-by: alshan <alshan@amazon.com>
Co-authored-by: Justin Kerr <justin.g.kerr@gmail.com>
* Necessary changes to establish baseline as reported by papers
- Port multiple train/eval split strategies from nerfstudio parser for colmap parser
- A number of fixes to establish solid baseline
- Avoid empty frames where no gaussians are projected onto the image
- Fix random order of image filenames
* a few enhancement changes to bridge gap between Inria and NS
* Fix PLY exporter and implement an PLY renderer
* Add finite filter in ply exporter
* Major changes to improve nerfstudio-gs training
* separate sh_0 and sh_rest and use different learning rates, and remove sh_rest regularization.
* disable downscale in warm-up stages
* prune the original gaussians that splits.
* change color clamp_max mechanism to be applied per ray instead of per gaussian
* Enable exporter and gsplat_render to work with color only PLY
* Prepare PR with minor changes
* Fix pyright errors
* remove gsplat ply renderer
* Address github/pwais comments
* fix an evaluation metric bug
* Use 0.1 cull_alpha_threshold for fast training
* Add scale regularization as an option (off by default)
* add continue_cull_post_densification option and some refactor
---------
Co-authored-by: Matthew Tancik <matt@lumalabs.ai>
* add orthophoto CameraType.
* test case of orthophoto camera.
* Fix the rotation errors caused by the usage of left-handed system
* Optimized ortho-cam, it can be used with other camera models now.
* fix TypeError of coords in `test_multi_camera_type()`
---------
Co-authored-by: Alexander Kristoffersen <akristoffersen@berkeley.edu>
* Adding docs for Instruct-GS2GS
Adding documentation for Instruct-GS2GS including its page and adding it to the list of external and 3rd party methods.
* fixing blank formatting dev checks
* updating subtitle
Updating the subtitle to specify gaussian splatting
* Add instructions for using Spectacular AI tools
Adds easy support for new device types: OAK-D, RealSense and
Azure Kinect, as well as an alternative pipeline for iOS.
* Grammar fixes in docs/quickstart/custom_dataset.md (Spectacular AI section)
Co-authored-by: Matias Turkulainen <30566358+maturk@users.noreply.github.com>
* Fix section label in the Spectacular AI section of custom_dataset.md
* Update custom_dataset.md: typo fix.
"montions" -> "motions"
---------
Co-authored-by: Matias Turkulainen <30566358+maturk@users.noreply.github.com>
fix pixel sampler setup for cameras with metadata other than fisheye crop radius
Co-authored-by: Matias Turkulainen <30566358+maturk@users.noreply.github.com>
* Add support for Project Aria datasets.
1) An export script for processing Project Aria datasets
2) Add support for Fisheye624 cameras
3) Add support for center-crop masking and sampling
Example commands for running nerfstudio on Project Aria data:
python scripts/datasets/process_project_aria.py --vrs_file=<path_to_vrs> --mps_data_dir=<path to MPS directory> --output_dir=<nerfstudio_data_dir>
ns-train nerfacto --data <nerfstudio_data_dir> nerfstudio-data --orientation-method none
* Clean up type errors
* Fix formatting
* Fix linter errors, import, and use absolute file paths
* Add aria subcommand to ns-process-data
* Appease ruff
* Add Aria to docs
* Fix docs, add back import try/except
* nit: path.as_posix() -> str(path)
---------
Co-authored-by: Brent Yi <yibrenth@gmail.com>
* Fix large dataset (> 512 images) training in viewer_beta (#2586)
* fix index bug in update_camera_poses of viewer
* Set Ruff line lenth
---------
Co-authored-by: Gina Wu <42229107+ginazhouhuiwu@users.noreply.github.com>
* Update hloc and add new matcher types
* Update dockerfile comment to better reflect what the command is doing
---------
Co-authored-by: mfischer <michael.fischer@spleenlab.ai>
* Fixes bug in PairPixelSampler when working with masked dataset
* Formatting
* Black formatting now ok.
---------
Co-authored-by: AdamRashid96 <71362382+AdamRashid96@users.noreply.github.com>
Co-authored-by: Brent Yi <yibrenth@gmail.com>
The ParallelDataManager (see #2092) makes pytorch crash if the cameras instance's fx, fy, cx or cy tensors
are loaded from a common shared tensor.
This PR fixes the issue by cloning the respective tensors before passing them to the Cameras(...) constructor.
