> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.stereolabs.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.stereolabs.com/_mcp/server.

# ZED SDK 3.1

> **Note**
>
> Release notes of the ZED SDK 3.1 series. Patch releases are listed newest first. Download links and the full changelog, including ZED Tools and integrations, are available on the [ZED SDK 3.1 release page](https://www.stereolabs.com/developers/release/3.1).

## What's New

ZED SDK 3.1 introduces new Camera control features such as Gamma adjustment for low-light environments and ROI (region of interest) selection for Auto Exposure/Gain.

Onboard sensors capabilities can now be accessed, including data such as sensors resolution, sampling rate, noise density, transforms, etc.

Depth map temporal stability and Quality mode have also been improved, along with many other bug fixes and improvements.

To update to 3.1 from previous 2.X versions, read the [migration guide](https://www.stereolabs.com/developers/release/3.0/migration-guide/).

## 3.1.2

### SDK

#### Improvements

* Greatly improved streaming and recording FPS on NVIDIA Jetson. Frame drops have been reduced from 15% to 1% when using HD720\@60fps with streaming or recording module enabled on Jetson Nano.
* Fixed a bug in H264/H265 SVO playback: The first frame was not correctly read and replaced by the second frame. This is now fixed and consistent with previous SDK versions.
* Added a security in object detection maximum perception distance. Depth maximum distance is now clamped at 100 meters to avoid memory overflow when using object detection.
* Added security and warnings when input parameters (`InitParameters`, `StreamingParameters`,...) are out of range.

## 3.1.1

### SDK

#### Improvements

* Fixed random jumps in ZED Mini IMU data. This was due to invalid timestamps that were sometimes generated.
* Fixed depth and normals maps in `SENSING_MODE::FILL` mode with confidence at 100. `NaN` or `Inf` values have been removed.
* Fixed a coordinate system issue in the object detection module. In certain coordinate systems, the distance between the left and right camera eyes was not set correctly, leading to invalid objects and position.
* Added a warning message when requested FPS is out of range (`HD2K` with fps=30 for example).
* Fixed a random camera opening issue that could return a wrong serial number and an `ERROR_CODE::INVALID_CALIBRATION_FILE` error in `open()`.
* Fixed required CUDA version declaration in the CMake file.
* Fixed Windows installer splash screen that could be displayed incorrectly.

## 3.1.0

### SDK

#### New Features

* Added new `VIDEO_SETTINGS::GAMMA` control to adjust gamma level of the camera. Can be very helpful in low light environments.
* Added new `VIDEO_SETTINGS::AEC_AGC_ROI` control to select a region of interest for auto exposure/gain adjustment. Exposure and Gain levels will be automatically optimized for the ROI and not for the full image. This is useful in high dynamic range scenes, for example when indoors and outdoors areas are visible at the same time. See `setCameraSettings(sl::VIDEO_SETTINGS,sl::Rect,sl::SIDE)` function for more information.
* Updated `CameraInformation` structure to improve availability of Camera, Sensors configuration and calibration information.
  * Added `SensorParameter` structure to access accelerometer, gyroscope, magnetometer, barometer configuration (resolution, sampling rate, noise density...).
  * Added `stereo_transform` (extrinsics stereo parameters) in user `COORDINATE_SYSTEM` and `UNITS`.
  * Updated `camera_imu_transform` to user `COORDINATE_SYSTEM` and `UNITS`.
* Added compression level option in `Mat::write()` function to adjust quality and size of PNG and JPG files.

#### Improvements

* Improved temporal stability and denoising of depth map.
* Improved accuracy of `QUALITY` depth mode.
* Improved performance when using the Object Detection module with the masks enabled.
* Jetson installation process has been improved. The installer now detects if it's running within Docker to skip any unnecessary steps. Object detection optimized AI model is now downloaded instead of being optimized on the host, making setup much faster.
* Improved cuDNN version handling.
* Added `Debug` compilation mode detection to warn about compatibility issues when mixing with Release libraries.

#### Bug Fixes

* Fixed `resetPositionalTracking()` that was not taking into account user-defined values for the yaw angle.
* Fixed Object detection AI model optimization that could lead to a corrupted model file. Additional checks have been added to handle file corruption.
* Fixed `getCameraSettings()` behavior when using Streaming input type. Now settings are available even if `setCameraSettings()` has not been called before.
* Fixed `getDeviceList()` on Windows. When multiple cameras were connected, the `DeviceProperties` of one of the cameras were wrong.
* Fixed multiple SVO recording in a single session. If `enableRecording()` was called again after `disableRecording()`, the SVO file was not correctly recorded.
* Fixed `setSVOPosition()` for H264/H265 SVO files that could return an image at an incorrect position.
* Fixed 3D Object Detection AI model optimization on Jetson Nano by increasing swap file size during the installation process.
* Improved robustness and reliability of the Object Detection AI model optimization process on all platforms.
* Improved compatibility with Darknet framework by adding `undef GPU` to solve the conflict with the macro and the SDK enumerate values.