v0.3

2024-04-05

Download for Windows

Changes

Training

• Added Preferences/Show Training Error to show PSNR (NeRF) or SSIM (Splat) in progress pane
• Support input image masking based on alpha channel also for Splat model
• Added option to assume black input background (experimental for synthetic datasets)
• Fixed occasional failure of camera tracker that was unrelated to the images
• Fixed failure to import COLMAP poses if 'sparse' folder contains unexpected files

Editing

• Added splat selection and deselection tool
• Added 'Invert Selection' action
• Added 'Delete Selected Splats' action
• Added 'Use Source Transform' button to restore original scene scale and orientation
• Crop box did not transform correctly if radiance field had been rotated

Import

• Support HEIF image format
• Fixed hang in some cases when importing variable frame rate video
  
• Fixed issues with file paths containing Unicode characters beyond Extended ASCII
• Fixed import failure when importing fewer poses than images
• Matching image file names to camera pose imports was too restrictive, leading to failed imports in some cases
• Fixed degenerate camera poses after re-orientation in some cases
• Camera tracking was pronounced failed too conservatively
• Removed upper limit on image resolution
• Prevent repeated error messages when image files are missing
• Image pre-processing progress did not show when importing poses
• Fixed an image selection issue when importing poses
• Ignore .db files (like Thumbs.db) for import instead of aborting import
• Importing images from network drives resulted in errors during tracking or camera pose import
• Fixed crash when importing and resizing RGBA images
• Abort import of camera poses if none match any imported images

Rendering

• Added Max Opacity parameter (aka Olli Huttunen's Ghost Wall parameter)
• Added option to render Z Depth
• Added Invert Depth and Max Depth parameters
• Rendered image/video files were cropped in some cases
• Fixed missing frames at the end of rendered video files in some cases
• Render to image sequence was broken

Export

• Apply scene transform and crop box to PLY exports

Camera Navigation

• Added Pan/Tilt/Roll Speed parameter in Scene/Viewport Camera

GUI

• Added Window/Reset Layout to revert changes to the UI tabs
• Update 'Elapsed' timer independently from progress bar
• Scale-to-fit rendered image if camera resolution is larger than viewport
• Don't popup message dialogs outside of main window
• Training progress bar did not activate in some cases
• Scene transform was not applied to sparse points in the viewport
• Fixed a crash when selecting ImageSet as viewport camera on large images
• Radiance field transform was not adjusted for when using Image Set as viewport camera
• Max Image Count UI did not update accrd. to selected Image Selection option
• Camera focal length was incorrectly changed when changing resolution
• Sparse points were temporarily drawn in black in some cases during camera tracking
• Some errors during import were not shown as popups (only in log)
• Use higher precision for training steps counter (for step counts > 1M)
• Increased maximum for Splat Scale from 4 to 10
• FOV was not updated when changing camera resolution
• Viewport camera did not update when deleting selected camera