XVFIOptions
Options & Configuration · AiDotNet.Video.Options
Configuration options for XVFI, extreme video frame interpolation
(Sim, Oh and Kim, ICCV 2021).
Inheritance: ModelOptions → NeuralNetworkOptions → XVFIOptions
For Beginners
XVFI handles very large motion in 4K video by estimating motion on a pyramid of ever smaller copies of the frames, starting at the coarsest, and refining it one level at a time. The same small network is reused at every level, so you can add levels at test time for bigger motion.
How It Works
The defaults are the release’s X4K1000FPS model (JihyongOh/XVFI: XVFInet.py, main.py, utils.py): 64 feature channels,
a 4x spatial reduction, three scale levels below the input in training and five at test; Vimeo returns
XVFI-v, the Vimeo90K model (2x reduction, one level). Training follows main.py: Adam at 1e-4, divided by 4 at epochs
100, 150 and 180, on the multi-scale L1 reconstruction loss plus half the edge-aware flow smoothness.
Constructors
| Constructor | Summary |
|---|---|
XVFIOptions() |
Initializes a new instance with default values. |
XVFIOptions(XVFIOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
OnnxModelOptions OnnxOptions { get; set; } |
Gets or sets the ONNX runtime options. |
double EdgeConstant { get; set; } |
Gets or sets the smoothness loss’s edge constant, squared in its exponent (150). |
double LearningRate { get; set; } |
Gets or sets Adam’s initial learning rate (1e-4). |
double LearningRateDecay { get; set; } |
Gets or sets the factor applied at each milestone (0.25). |
double ReconstructionWeight { get; set; } |
Gets or sets the weight of the multi-scale reconstruction loss (rec_lambda, 1). |
double SmoothnessWeight { get; set; } |
Gets or sets the weight of the edge-aware flow smoothness loss (0.5). |
int NumFeatures { get; set; } |
Gets or sets the feature channels (nf, 64). |
int ScaleFactor { get; set; } |
Gets or sets the feature extractor’s spatial reduction, a power of two (module_scale_factor, 4). |
int TestDepth { get; set; } |
Gets or sets the lowest scale level at test (S_tst, 5). |
int TrainingDepth { get; set; } |
Gets or sets the lowest scale level in training (S_trn, 3). |
int[] LearningRateMilestones { get; set; } |
Gets or sets the epochs at which the rate is multiplied by LearningRateDecay (100, 150, 180). |
string ModelPath { get; set; } |
Gets or sets the path to the ONNX model file. |
Methods
| Method | Summary |
|---|---|
XVFIOptions Vimeo() |
XVFI-v, the release’s Vimeo90K configuration: a 2x reduction and one scale level in training and test. |