Annotate individual frames with one-click oriented cuboids, inspect each one in top, front and side views, view the camera image that automatically matches your viewport direction, project cuboids directly onto camera images, assign persistent tracking IDs, and propagate annotations across adjacent frames instantly via the Tracking ID Manager.
Eliminate manual frame-by-frame labeling. By placing bounding cuboids at the start and end anchor frames, Sycrad automatically calculates object positions, headings, and persistent identities across all intermediate frames.How Interpolation works ↘
| Dataset | Objects | Frames | IoU ≥ .7 |
|---|---|---|---|
| KITTI | 36 | 1098 / 1098 | 93% |
| nuScenes | 55 | 354 / 359 | 75% |
| Waymo | 62 | 6364 / 6434 | 96% |
Register a full frame range into a single unified scene. Instead of discarding moving objects to prevent ghosting, Sycrad detects and tracks dynamic objects over time, automatically generating labeled cuboids that distribute back to individual source frames in one click.How SLAM works ↘
| Dataset | Matched | FP | FN | Precision | Recall | IoU |
|---|---|---|---|---|---|---|
| KITTI | 660 | 14 | 118 | 97.9% | 84.8% | 0.72 |
| nuScenes | 254 | 81 | 139 | 75.8% | 64.6% | 0.68 |
| Waymo | 2612 | 272 | 1135 | 90.6% | 69.7% | 0.71 |
Objects indicates cuboids labeled across anchor frames; Frames measures generated bounding boxes against expected frame count; IoU ≥ .7 reflects cuboid overlap meeting the 70% accuracy threshold. Matched pairs generated predictions with ground-truth objects; FP denotes false positive detections; FN counts missed true objects; Precision represents the valid share of all predictions; Recall reflects the recovered share of true objects; and IoU indicates the mean volumetric overlap across matched bounding cuboids.
See single-frame annotation, interpolation, SLAM, and aerial visualization inside the same native workspace.
Built for absolute data sovereignty. By processing point clouds locally, Sycrad eliminates both data exposure risks and recurring cloud transfer fees. Cloud services are used for license validation, while Enterprise plans run fully air-gapped with zero network requirements.
Qt's QRhi interfaces directly with native graphics APIs to eliminate runtime overhead. Sycrad handles level-of-detail and visibility upfront, maximizing GPU throughput on dense scenes.
QRhi
Metal · Direct3D 12
Vulkan in progress
RTX 4070 SUPER benchmark
218M
points
240
FPS
~30%
GPU load
Near-idle
CPU
4.3 GB
VRAM
4.6 GB
RAM
Performance varies with the view, settings, dataset, and hardware.
Dynamic distance-adaptive level of detail preserves fine structural clarity near the camera while minimizing GPU vertex overhead on distant points.
Discards point-cloud nodes outside the active field of view in real time, keeping render throughput focused strictly on visible scene geometry.
Engineered as a native C++ desktop platform, free from browser memory limits and web runtime constraints.
Distributes computational workloads and data processing across multi-core processors using oneTBB for maximum throughput.
