Type Definitions

Video Effects SDK type definitions provide selector strings for video effect filters and the parameters of a video effect filter.

NvVFX_EffectSelector

typedef const char* NvVFX_EffectSelector;

This type definition provides the selector strings for the various types of video effect filters.

NVVFX_FX_TRANSFER “Transfer”

Image transfer effect.

This effect provides the same capability as the NvCVImage_Transfer() function in the form of an effect. This effect is especially useful to match formats in a pipeline of effects.

NVVFX_FX_GREEN_SCREEN “Green Screen”

AI green screen filter.

NVVFX_FX_BGBLUR “Background Blur”

Background Blur filter.

NVVFX_FX_SUPER_RES “Video Super Res”

AI-based video super resolution.

NVVFX_FX_SR_UPSCALE “Upscale”

AI-based fast video upscaler.

NVVFX_FX_DENOISING “Denoising”

Webcam Denoising filter.

NVVFX_FX_VIDEO_FRAME_GENERATION “VideoFrameGeneration”

AI-based temporal frame interpolation filter. For details about its operating modes and parameters, refer to Video Frame Generation.

NVVFX_FX_TRUE_HDR “TrueHDR”

AI-based SDR-to-HDR conversion filter. For details of the tunables that this filter accepts, refer to TrueHDR.

NvVFX_ParameterSelector

typedef const char* NvVFX_ParameterSelector;

This definition type provides the selector strings for the parameters of a video effect filter.

NVVFX_INPUT_IMAGE_0 “SrcImage0”

The NvCVImage structure that will be used as the input to the effect.

Because no effect takes more than one input image, this selector is equivalent to NVVFX_INPUT_IMAGE.

NVVFX_OUTPUT_IMAGE_0 “DstImage0”

The NvCVImage structure that will be used as the output of the effect.

Because no effect has more than one output image, this selector is equivalent to NVVFX_OUTPUT_IMAGE.

NVVFX_MODEL_DIRECTORY “ModelDir”

The path to the folder that contains the model files that will be used for the transformation.

NVVFX_MODEL_CACHE_DIRECTORY “ModelCacheDir”

(Windows on Arm only, such as RTX Spark) Optional path to the model runtime cache directory. Set this parameter before calling NvVFX_Load(). If it is empty or is not set, the SDK uses a cache subdirectory under ModelDir. Applicable to AI Green Screen, Webcam Denoising, and Video Relighting.

NVVFX_MODEL_CACHE_MODE “ModelCacheMode”

(Windows on Arm only, such as RTX Spark) Optional unsigned integer that controls the model runtime cache. Set this parameter before calling NvVFX_Load():

• 0: Auto (default). Load a valid cache if present; otherwise run JIT and save a new cache.

• 1: Disabled. Do not read or write cache files.

• 2: Force regenerate. Ignore any existing cache, run JIT, and write a fresh cache.

On the first load of each model without a valid cache, JIT compilation can require on the order of several seconds. Later loads reuse the cache when the mode is Auto. Applicable to AI Green Screen, Webcam Denoising, and Video Relighting.

NVVFX_CUDA_STREAM “CudaStream”

The CUDA stream in which to run the video effect filter.

NVVFX_CUDA_GRAPH “CudaGraph”

Enable CUDA Graph optimization to reduce GPU operation submission overhead. Enabling this flag is recommended on Windows on Arm (for example, RTX Spark) for better steady-state performance where the effect supports it.

NVVFX_INFO “Info”

Get information about a video effect filter and its parameters.

NVVFX_MAX_INPUT_WIDTH “MaxInputWidth”

Maximum width of the supported input.

NVVFX_MAX_INPUT_HEIGHT “MaxInputHeight”

Maximum height of the supported input.

NVVFX_MAX_NUMBER_STREAMS “MaxNumberStreams”

Maximum number of concurrent input streams.

NVVFX_SCALE “Scale”

Scale factor. This is used to scale the values of the images during transfer to match formats in a pipeline of effects.

NVVFX_STRENGTH “Strength”

Strength for the filters that use this parameter. Higher strength implies a stronger effect. For Video Super Resolution, use a floating-point value in the interval [0.0, 1.0] to control the intensity of the selected mode.

NVVFX_STRENGTH_LEVELS “StrengthLevels”

Number of unique strength levels in the interval [0, 1]. Currently this applies only to the Webcam Denoise filter and is set to 2, which implies that the two strength levels are 0 or 1.

NVVFX_MODE “Mode”

The mode selector is used by filters that expose multiple operating modes. For Video Super Resolution mode values, refer to Video Super Resolution.

For AI Green Screen, the values are as follows:

• 0: Quality mode with chairs segmented as the foreground

• 1: Performance mode with chairs segmented as the foreground

• 2: Quality mode with chairs segmented as the background

• 3: Performance mode with chairs segmented as the background

NVVFX_TEMPORAL “Temporal”

Apply temporal filtering.

NVVFX_GPU “GPU”

Preferred GPU to use. This is an optional parameter.

NVVFX_BATCH_SIZE “BatchSize”

Size of a batch of input provided to a filter. The default value is 1.

NVVFX_MODEL_BATCH “ModelBatch”

The preferred batching model to use, which is tuned for a batch size of 1 or 8. This is applicable only to the AI Green Screen filter.

NVVFX_STATE “State”

An array of state variables.

NVVFX_STATE_SIZE “StateSize”

The number of bytes needed to store the state variable.

NVVFX_STATE_COUNT “NumStateObjects”

The number of active state object handles.