SLAMForge Desktop 3.2.0-beta.2

SLAMForge Desktop 3.2.0-beta.2

This beta aligns the Windows video-processing runtime with the validated Linux configuration. It
does not introduce a new SLAM algorithm; it removes known dependency and decoder differences so
the same video and calibration can be compared more meaningfully across platforms.

Downloads

• SLAMForge-Desktop-3.2.0-beta.2-Windows-x64.zip — portable Windows 10/11 x64 package.
• SLAMForge-Desktop-3.2.0-beta.2-Linux-x86_64.AppImage — portable Linux x86_64 application.
• SHA256SUMS.txt — checksums for both packages.

Windows users must extract the complete ZIP before launching SLAMForge Desktop.exe. The package
is unsigned, so SmartScreen may display an unrecognized-publisher warning.

What changed

• The Windows package now uses OpenCV 4.6.0 with the compatible FFmpeg 4.4.3 dependency instead of
  resolving a newer OpenCV and an unrelated decoder stack.
• Video input explicitly prefers FFmpeg on Windows and Linux, with automatic fallback only when
  FFmpeg cannot open the source.
• slamforge_cli --version and each run header report the exact SLAMForge, OpenCV, Ceres, compiler,
  and decoder-backend information.
• Windows desktop and release CI reject packages that do not contain the validated OpenCV 4.6.0
  runtime.

How to compare Windows and Linux

Use the same original video and the same calibrated YAML file on both systems. Keep all processing
settings identical. In each run.log, confirm that the header reports OpenCV 4.6.0 and the video
decoder reports FFMPEG before comparing trajectory.txt, tracking-loss intervals, and the PLY
outputs.

Floating-point results from MSVC and GCC are not guaranteed to be byte-identical. This release
removes the known OpenCV and decoder mismatch; it does not claim that the custom SLAM core is SOTA
or that every remaining platform-dependent numerical branch has been eliminated.

Existing limitations

• Monocular reconstruction has unknown absolute scale and requires accurate camera calibration.
• Dense output is a fused surface point cloud, not a watertight mesh or survey-grade model.
• The downloadable package is CPU-only and may take substantially longer than the video duration.
• Optional g2o/FBOW acceleration backends are not included in the desktop packages.

SLAMForge is GPL-3.0-only. Depth Anything V2 Small and its ONNX conversion are Apache-2.0; ONNX
Runtime is MIT. The differently licensed Depth Anything Base/Large/Giant weights are not included.

SLAMForge Desktop 3.2.0-beta.1

SLAMForge Desktop 3.2.0-beta.1

This is the first SLAMForge desktop beta that reconstructs a dense colored surface instead of only
displaying sparse ORB landmarks. Select or drop a video, provide the matching camera calibration,
and the packaged application performs tracking, loop correction, depth inference, multi-view
filtering, and fusion locally.

Downloads

• SLAMForge-Desktop-3.2.0-beta.1-Windows-x64.zip — portable Windows 10/11 x64 package.
• SLAMForge-Desktop-3.2.0-beta.1-Linux-x86_64.AppImage — portable Linux x86_64 application.
• SHA256SUMS.txt — checksums for both packages.

Windows users must extract the complete ZIP before launching SLAMForge Desktop.exe. The package
is unsigned, so SmartScreen may display an unrecognized-publisher warning.

What is new

• Bundled Depth Anything V2 Small and ONNX Runtime; dense inference needs no Python, network access,
  GPU, or separate model download.
• Learned depth is scale/shift aligned to geometric SLAM landmarks in each selected keyframe.
• Adjacent-view consistency checks and deterministic colored voxel fusion remove unsupported depth.
• The result viewer renders the actual scene colors and dense surfaces.
• Results now contain map.ply (dense), sparse_map.ply (landmarks), trajectory.txt, and
  run.log.
• Weak extreme-scale loops can no longer apply the 20–30x route deformation seen in beta.2.

Expectations and limitations

• Processing has a second offline dense stage and therefore takes longer than beta.2.
• A monocular camera cannot determine absolute metric scale; all geometry remains relative scale.
• Accurate intrinsics and distortion parameters are still required.
• Dense learned depth makes walls and large objects continuous and recognizable, but it is not
  survey-grade geometry and can be unreliable on mirrors, transparent objects, extreme blur, or
  imagery far outside the model's training distribution.
• map.ply is a surface point cloud, not a watertight triangle mesh. Monocular depth and pose drift
  can still produce thick walls, duplicated edges, floating layers, or gaps.
• The downloadable package is CPU-only. A discrete GPU is not required.

Release validation

Two complete runs of the 4,757-frame rectified reference video produced byte-identical trajectories
and sparse maps. Each run exported 4,556 poses with no post-initialization lost frames, 598
keyframes, two rigid loop corrections, and an 886,813-point dense surface cloud. On the 954 poses
with finite ground truth, global Sim(3) alignment gave 1.554 m ATE RMSE. These numbers are a
single-sequence regression check, not a claim of survey accuracy or broad benchmark performance.

SLAMForge is GPL-3.0-only. Depth Anything V2 Small and its ONNX conversion are Apache-2.0; ONNX
Runtime is MIT. The differently licensed Depth Anything Base/Large/Giant weights are not included.

SLAMForge Desktop 3.1.0-beta.2

SLAMForge Desktop 3.1.0-beta.2

This beta packages the rewritten SLAM tracking and mapping pipeline used by the current SLAMForge
source. Users of 3.1.0-beta.1 should replace the complete extracted application directory; the old
Windows ZIP does not contain these algorithm changes.

Downloads

• SLAMForge-Desktop-3.1.0-beta.2-Windows-x64.zip — portable Windows 10/11 x64 package.
• SLAMForge-Desktop-3.1.0-beta.2-Linux-x86_64.AppImage — portable Linux x86_64 application.
• SHA256SUMS.txt — checksums for both packages.

SLAM reliability update

• Robust PnP, optical-flow-assisted tracking, essential-matrix recovery, and recovery after
  temporary tracking loss.
• Safer keyframe creation, multi-view triangulation, pose synchronization, and deterministic
  offline local mapping.
• Stable Ceres local bundle adjustment that excludes underconstrained landmarks and invalid camera
  projections.
• Multi-loop geometric verification and Sim(3) drift correction, with a deterministic fallback that
  works in the portable packages without FBOW or g2o.
• Stable trajectory export and sparse-map outlier filtering.

On the 4,757-frame reference video used for this update, beta.2 exports 4,554 poses and reconstructs
the complete closed route without the repeated dense-Cholesky failures observed with beta.1.

Desktop workflow

• Drag or select MP4, MOV, AVI, MKV, and M4V video files.
• Select a calibrated SLAMForge camera YAML and writable result directory.
• Run or cancel the isolated SLAM worker while monitoring progress and logs.
• Inspect the completed sparse map and camera trajectory in the interactive result view.
• Export ASCII PLY, TUM/KITTI/EuRoC trajectory output, and a persistent run log.
• Process locally without uploading the selected video or generated map.

Important limitations

• Accurate camera calibration is required and is not inferred automatically.
• Monocular reconstruction has unknown absolute scale.
• Output is a sparse landmark map and camera path, not a dense surface model.
• Low texture, blur, pure rotation, repeated patterns, or insufficient parallax can still degrade
  initialization and tracking.
• Optional FBOW/g2o acceleration backends are not bundled; the built-in geometric loop-closing
  fallback remains enabled.
• The Windows beta is unsigned and may trigger a SmartScreen warning.

SLAMForge is distributed under GPL-3.0-only. Please report reproducible beta issues with the video
metadata, camera configuration, run.log, operating system, and exact package filename.

SLAMForge Desktop 3.1.0-beta.1

SLAMForge Desktop 3.1.0-beta.1

This is the first downloadable SLAMForge Desktop beta. It packages the SLAM engine and Qt desktop
interface so users can process videos without building the source code.

Downloads

• SLAMForge-Desktop-3.1.0-beta.1-Windows-x64.zip — portable Windows 10/11 x64 package.
• SLAMForge-Desktop-3.1.0-beta.1-Linux-x86_64.AppImage — portable Linux x86_64 application.
• SHA256SUMS.txt — checksums for both packages.

What works

• Drag or select MP4, MOV, AVI, MKV, and M4V video files.
• Select a calibrated SLAMForge camera YAML and writable result directory.
• Run or cancel the isolated SLAM worker while monitoring progress and logs.
• Display the completed sparse map and camera trajectory in an interactive result view.
• Export ASCII PLY, TUM/KITTI/EuRoC trajectory output, and a persistent run log.
• Local-only processing with no video or map upload.

Important limitations

• Accurate camera calibration is required and is not inferred automatically.
• Monocular reconstruction has unknown absolute scale.
• Low texture, blur, pure rotation, repeated patterns, or insufficient parallax can prevent
  initialization or tracking.
• Live per-frame map rendering is not included yet; the result viewer updates after processing.
• The packages enable Ceres local bundle adjustment but omit optional g2o/FBOW loop closing.
• The Windows beta is unsigned and may trigger a SmartScreen warning.

SLAMForge is distributed under GPL-3.0-only. Please report reproducible beta issues with the video
metadata, camera configuration, run.log, operating system, and package filename.