v0.4

2024-08-23

Download for Windows

Changes

After Effects

Fixed: render failed after recent change (black output)

Camera

Fixed crash when trying to use cameras with degenerate/extreme distortion
Pose import from Reality Capture did not handle Brown4 models correctly
Fixed: aspect ratios < 1 (portrait mode) did not render correctly
Fixed a bug that prevented camera editing/animation
Added File/Export Cameras
Added 'Track Target' parameter to improve camera animation workflow and interpolation smoothness

Command-line interface

Added postshot-cli.exe to allow training from the command line
See Command-line interface for more details
Don't require a display GPU
Try to save file before training to detect failure early
Prevent hang in some cases when importing images only (no camera poses)
Added --show-train-error flag
--max-num-features was not interpreted in units of 1000 as the equivalent GUI parameter
Fixed a diagnostic error that was printed in some cases
Profile selection was broken, leading to training not starting correctly
Terminating using Ctrl+C keyboard interrupt caused crash during shutdown in some cases
Support login from the command-line to allow operation on (non-graphics) server GPUs
Added --gpu option to select used GPU
Fixed: --max-image-size was not applied
Added default values for --max-num-features and --max-image-size to match GUI import

Framework

Moved app data files %USERPROFILE%\PostshotData to %LOCALAPPDATA%\Postshot
Fixed: too much GPU memory was kept when closing scenes
Cleaned up some redundant log messages
Don't retain old log files
Workaround an AMD driver bug that prevented Postshot from starting on some mixed-GPU systems
Fixed unsupported column 'a' error when saving projects with some types of imported point clouds
Fixed an error that caused renders to fail in some cases
Fixed: hang during 'Optimizing GPU Code' on some systems unless run as adminsitrator

Image

Support RAW file formats
Support 16bit and 32bit integer colors
New image streaming system supports image sets that don't fit into CPU memory
Downscale on-demand during Rdnc Field training io. during pre-processing
Prevent excessive CPU memory usage when selecting images from long videos/image sequences.
Abort training only if all images or cameras are missing (as opposed to only one)
Store Postshot version in EXR attributes
Fixed a crash in some cases when importing an incomplete pose set
Fixed: image pre-processing was run when only loading project in some cases
Fixed failure to store projects in some cases ("duplicate chunk name" error)
Added image set name to log for several operations
Fixed a memory leak when training on images with mixed pixel formats
Fixed: downscaling did not work correctly on EXR images
Fixed: rendering to image sequence failed to save several frames in some cases

Import

Support image sets with multiple camera intrinsics
Support importing COLMAP data in .txt format
Support import of Reality Capture cameras/points in CSV/PLY formats
See Importing Images for more details.
Support Bundler (*.out) file format to importing poses and points
Set Frame Selection to 'Use All Images' when importing poses
Increased maximum recommended 'Max Image Count' to 1000
UI: added file list and allow dropping files multiple times
UI: adjust default Train Steps Limit value to image count
Fixed: importing cameras from .json implicitly imported images

Installer

Added 'Accept EULA' checkbox

NeRF

Result was broken if input image downscaling was used

Nerf pipeline did not handle display orientation in input videos correctly

Render

Show error if render output path is missing extension io. silently not storing output
Added Camera/Tonemapping option
Disabled far clipping
Fixed hang when rendering without radiance field
Fixed crash in renderer accompanied by Vulkan device lost errors
Fixed: render did not restrict render resolutions to window size
Fixed: Poses on Camera objects created from Image Sets were incorrect

Scene

Support render camera imports from Alembic files
Retain scene transform from tracked cameras by default

SfM

Added 'Single Lens & Focal Length' tracking parameter
Fixed an issue that could cause the camera tracker to abort erroneously
Fixed issue that causes distorted cameras/scene for some image sets
Fixed crash when aborting camera tracking during Step 3/4
Fixed issue that caused camera tracker to abort in some cases on images >4k resolution.
Fixed: pausing/aborting CameraTracker did not work during some stages.

Splat

Added Sky Model option
Support training on HDR images
Support color spaces other than sRGB incl. ACES, HLG/Rec. 2020, Adobe RGB
Added new 'Splat MCMC' profile with different densification behavior
Some model quality improvements
Reduced GPU memory usage during training by ~15%
Several training performance and memory management optimizations
Use original color space when rendering to EXR
Added Render/Exposure parameter
Added Rdnc Field/Render/Min Radius parameter
Release more pre-allocated memory when closing a project.
Fixed an issue that caused artefacts similar to overexposure
Fixed: training/rendering images larger than 8K did not work
Fixed: training did not abort if images failed to load
Fixed: selected splats could get 'stuck' when storing project with active selection
Fixed a crash when saving projects with an empty radiance field
Fixed: on multi-GPU systems, training on GPUs other than GPU 0 did not work
Fixed: Switching between Splat and NeRF profiles on existing project was broken
Fixed a crash when loading projects stored with older versions

UI

Added separate brush selection mode without depth limit

Added option to relocate missing images/videos

Added ImageSet/Timeline Selects Image option

Added ImageSet/Create Camera action to allow rendering training cameras

Added checkbox in import dialog to quickly disable image downsample size

Added 'File/Open Recent' submenu. Don't auto-load last project on startup.

Added shift key modifier for fast camera movement using WASDQC shortcuts.

Added Downsample parameter to RdncField node iot. allow change after import.

Added parameter for Home Camera transform in Scene/Parameters and Ctrl+Shift+H shortcut to set it

Added Help/About dialog

Always display import dialog centered io. following mouse

Rdnc Field: show trained step count in Model Stats

Show number of registered images during camera tracking.

Image Set: added separate image counts for imported, selected and tracked image sets

Image Set: moved action buttons to separate parameter tab

Show last avg. training error in Rdnc Field/Model Stats after training

Replaced obsolete 'Use Scene Transform' with 'Align Up Vector'

Mouse wheel changes selection brush size faster

Mouse wheel changes depth limit faster

Made selection more sensitive to include partially hovered splats

Use scientific notation for very small or large numbers (1e-6 etc.)

Changed shortcuts for selection tool to avoid collision with camera navigation (see status message for details)

Increased the default camera move speed in Ego mode

Increased tab dragging threshold to help prevent accidental layout changes

Fixed: Progress UI did not abort in some cases when Camera Tracker aborted

Fixed: training did not start (after camera tracking) if window was minimized

Fixed: don't load image files when using Image Set only as viewport camera on pre-trained model

Fixed: Rdnc Field/Model Stats splat count was incorrect in some cases

Fixed: Preferences/Show Train Error was not correctly restored from stored settings

Fixed: Login failed on some systems

Fixed: Viewport camera was not restored correctly in some cases when loading a scene

Fixed: Progress tab remained inactive during image pre-processing stage on any but the first job

Fixed: Dragging number fields created many undo steps io. grouping into one

Fixed: Import file selector was not listing some supported file formats

Fixed: Changing resolution on Camera Node updated horizontal FOV io. vertical only

Fixed: Splat Density was not restored correctly when loading projects

Fixed: UI did not update correctly after display/system sleep

Fixed: camera pose widgets were drawn corrupted in some cases

Fixed: renaming a selected camera object switched the viewport to the default camera

Fixed: progress timers kept counting while training was paused

Fixed: using render action with no camera object switched viewport to Image Set camera

Fixed: Model Stats field did not update when resetting radiance field

Fixed: incorrect error message when starting render while no camera was selected

Fixed a UI hang when creating a new project in some cases

Fixed: Pausing/aborting training during 'Loading Images' stage did not work.

Fixed: Viewport was not cleared when switching to camera with different aspect ratio.

Fixed: main menu popups did not close in some cases when clicking outside of menu.

Fixed: Some Unicode characters in recent file paths caused error on startup

Fixed: camera poses and points did not update during camera tracking

Fixed: File/Open Recent list was restored in reverse order

Fixed a crash when using Windows/Reset Layout

Fixed: Import dialog did not show unit symbol on Max Splat Count parameter

Fixed: Allow entering invalid values into number fields, revert when exiting field instead

Fixed an issue causing icons to jitter in some cases

Fixed: Step-Limit-Reached message did not popup correctly in some cases

Fixed: dragging number elements caused extra undo entry after recent change

Fixed: timeline jumped to incorrect frames in some cases

Fixed: File/Open did not reset timeline correctly

Fixed overly sensitive mouse controls when using UI via Remote Desktop

Fixed incorrect 'Remaining' time in progress widget when continuing a training job

Fixed: importing images and COLMAP files gave incorrect error about missing points

User Guide

Added Importing Images
Added Command-line Interface