Use ns-process-data images --no-same-dimensions ... to handle images sets with different dimensions properly. (Fast downscaling assumes image dimensions are the same.)
* Remove system-level CUDA install from Linux install instructions
* Remove system-level CUDA from Windows install instructions
* Un-flip CUDA 11.7 / 11.8
* Fix pixel sampler cpu bottleneck
cpu usage reduced 1700% -> 100%, and even more if using multi-gpu training
* Convert to config option
* Change optional to bool
* Moved Volinga as an external method
* Update external_methods.py for black
* Formatted with black
---------
Co-authored-by: jperezh <jperezh@arquimea.com>
* Render VR180 video
Adding the VR180 camera model to render VR180 video and images. I condensed the omnidirectional stereo and VR180 render post processing in the render script. I have updated the documentation for VR180 support.
* Condensing formatting of camera types
Condensing the if statement for camera type intrinsics.
* Implement nerfacto alpha transparency training
* Stop adding all ones alpha channel and clean up blend_background
* Make background_color a required argument in blend_background
* Enable alpha transparent training by default
* Improve background_color typing
* Blend uniform background color for metric computation
* Fix typing error
* Clean up typing
* Fix background_color to outputs assignment
* Model outputs returns unblended RGB (black bg) for random bg
* Update renderers
* Fix comments
* Add support for transparent BG for other methods
* Minor refac
* Fix get_background_color
* Fix typing error for blend_background
* Fix comments
* Export utils support transparency
* Pylint fix
* Fix pcd export masking
* Filter pcd view directions based on opacity
* Clean up get_rgba_image
---------
Co-authored-by: Jonas Kulhanek <jonas.kulhanek@live.com>
* Fix arguments that are not passed from config to NerfactoField
* Add comments
* Reset base_res and features_per_level config of nerfacto-big and nerfacto-huge to default to match the original behavior
---------
Co-authored-by: Matthew Tancik <matt@lumalabs.ai>
* Merge video image extraction and downscaling
* Linting
* Fix extracting all frames, only apply vfr when thumnailing
* integrate downscale chains into image processing
* fast polycam upscale/downscale depth
* black
* black
* Update process_data_utils.py
Remove code change from other branch.
* use same suffix for downscale, don't force png
* black
---------
Co-authored-by: Sebastiaan Meijer <seb@stiaan.io>
* Try non-strict model loading on RuntimeError. ...
This allows tcnn models that contained a torch hash_table to load.
In particular, it allows us to load tcnn models prior to PR #2175.
cf. https://github.com/nerfstudio-project/nerfstudio/pull/2175
* Only allow non-strict loading when called with defaults.
* Reflow to avoid code duplication; blacken.
---------
Co-authored-by: Matthew Tancik <matt@lumalabs.ai>
Update to omnidirectional stereo rendering (VR 360) to fix slight positional alignment to accurately composite with Blender objects. Modified documentation for ODS and the Blender add-on to include details on Blender camera settings to composite with the Nerfstudio ODS camera model.
