> For the complete documentation index, see [llms.txt](https://docs.supervisely.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.supervisely.com/import-and-export/import/supported-annotation-formats/pointcloud_episodes/seyond_pcap.md).

# Seyond lidar .pcap

## Overview

{% hint style="success" %}
Import Seyond lidar captures (raw `.pcap` files) as point cloud episodes: each sensor's recording becomes an episode.
{% endhint %}

## Format description

**Supported point cloud format:** `.pcap` (classic libpcap, as written by `tcpdump`; not pcapng)\
**Supported lidars:** Seyond Falcon (sphere point cloud packets). Captures from other Seyond models, such as Robin W, are recognised and skipped with a message.\
**Supervisely version:** 6.18.14 or later; Import as Links from cloud storage needs 6.18.17\
**With annotations:** only labels exported by the **Export Seyond lidar pcap** app\
**Data structure:** Information is provided below.

A capture is the lidar's own UDP output recorded to a file, usually one file per sensor every minute. Supervisely does not convert it. The capture is stored once, as an upload, a link to your cloud storage, or a file in Team Files, and every lidar frame in it becomes an item that reuses the file. The labeling tool decodes each frame from the capture.

Each frame gets:

* `meta.frame`: its frame number
* `meta.timestamp`: the capture time of the frame's first packet (UTC). The episode player shows time from it
* `meta.sensorsData.extrinsicMatrix`: the sensor's mount, when `sensors.json` gives it

Frames are numbered in the order they start in the file. The first and last frames of a capture are usually partial sweeps, and they are kept.

## Input files structure

Both directory and archive are supported, and so are links to cloud storage (**Import as Links**).

```
📦captures (folder or .tar/.zip archive)
├──📄sensors.json (optional)
├──📄truck-1_forwardRight_falcon_2026-08-26--16-48-00.pcap
├──📄truck-1_forwardRight_falcon_2026-08-26--16-49-00.pcap
├──📄truck-1_forwardLeft_falcon_2026-08-26--16-48-00.pcap
└──📄...
```

Captures named `{device}_{location}_{model}_{timestamp}.pcap` are grouped by sensor (`{device}_{location}`). A capture with any other name is its own sensor, named by its file name without the extension.

### Sensor mount: `sensors.json`

A capture does not record how the lidar is mounted. Give each sensor's mount (sensor to world) so the points, and the labels drawn on them, are in vehicle axes. A mount cannot be changed once a frame has labels, so set it at import.

```json
{
  "sensors": {
    "truck-1_forwardRight": {"xyz": [1.2, -0.8, 2.1], "rpyDegrees": [-90, 0, 0]},
    "truck-1_forwardLeft": [1, 0, 0, 0, 0, 0, 1, 0, 0, -1, 0, 0]
  },
  "episodes": {"maxGapSec": 1.0, "maxFrames": 3000}
}
```

* A mount is 12 numbers (a 3×4 row-major matrix), 16 numbers (4×4 with the last row `0 0 0 1`), or `xyz` in metres with `rpy` in radians or `rpyDegrees`. The rotation is `Rz(yaw) · Ry(pitch) · Rx(roll)`.
* It must be a rotation plus a translation, with no scale or mirroring.
* Without a mount, the import still runs and warns that the points stay in the sensor's own axes.

## What the import creates

* The captures of each sensor are chained by time into episodes. The next capture continues an episode when it starts within `episodes.maxGapSec` (1 second by default) of the previous one, and while the episode stays within `episodes.maxFrames`. A capture is never split.
* Each episode is a dataset named after its sensor, with the start time added when a sensor has several. A single episode goes into the dataset you import into.
* One frame per lidar frame, named `<capture name>_0000.pcap` and so on. `meta.frame` continues across the captures of an episode.
* Where two captures meet, the lidar sweep that was cut by the file rotation shows as two partial frames, one from each file.

## Export

The **Export Seyond lidar pcap** app exports the captures (each once, byte-identical), the sensors' mounts and the episode labels. Importing that folder again restores the labels on the same frames.
