InterpAnyClearerOptions

Options & Configuration · AiDotNet.Video.Options

Configuration options for InterpAnyClearer, distance indexing and iterative reference-based estimation for clearer anytime frame interpolation (Zhong et al., ECCV 2024).

Inheritance: ModelOptions → NeuralNetworkOptions → InterpAnyClearerOptions

For Beginners

Ordinary interpolation is told only “halfway in time”, which is ambiguous when objects speed up or slow down, so the result blurs. This model is instead told, for every pixel, how far along its path the object should be (the distance index). It reaches that position in a few small steps, each step starting from the frame the previous step produced, which keeps the result sharp. A map that is the same everywhere behaves like a time.

How It Works

The defaults are the release’s DR-RIFE (zzh-tech/InterpAny-Clearer, models/DI-RIFE: IFNet_sdi_recur.py, refine.py, RIFE_sdi_recur.py, train_sdi_m_mask_recur.py): RIFE’s IFBlocks of widths 240, 150 and 90 at scales 4, 2 and 1, a width-90 privileged teacher, the width-16 Contextnet and U-Net, each block reading the five-channel index map (target distance, reference distance, reference frame), and two reference iterations at inference. Training follows the release: AdamW with a weight decay of 1e-3, the rate rising linearly to 3e-4 over 2,000 steps and falling by a cosine to 3e-6 over 300 epochs, on the Laplacian loss of the frame and of the teacher plus 0.002 times the distillation loss.

Constructors

Constructor Summary
InterpAnyClearerOptions() Initializes a new instance with default values.
InterpAnyClearerOptions(InterpAnyClearerOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets the ONNX runtime options.
bool UseDistillation { get; set; } Gets or sets whether the teacher block exists and distills the students in training (true).
double DistillationWeight { get; set; } Gets or sets the distillation loss’s weight (0.002, the continuous-index setting).
double LearningRate { get; set; } Gets or sets AdamW’s peak learning rate, reached after the warmup (3e-4).
double MinLearningRate { get; set; } Gets or sets the rate the cosine decays to (3e-6).
double WeightDecay { get; set; } Gets or sets AdamW’s decoupled weight decay (1e-3).
int ContextChannels { get; set; } Gets or sets the Contextnet’s and U-Net’s base width c (16).
int Iterations { get; set; } Gets or sets the reference iterations from frame 0 to the target index map at inference (iters, 2).
int LaplacianLevels { get; set; } Gets or sets the Laplacian loss’s pyramid levels (5).
int TeacherWidth { get; set; } Gets or sets the privileged teacher IFBlock’s width (90), even.
int TotalSteps { get; set; } Gets or sets the step at which the cosine reaches MinLearningRate: 300 epochs of the septuplet training set at the release’s 1,009 steps per epoch (302,700).
int WarmupSteps { get; set; } Gets or sets the steps over which the rate rises linearly from zero (2,000).
int[] BlockWidths { get; set; } Gets or sets the three student IFBlocks’ widths, coarsest first (240, 150, 90); each is even.
string ModelPath { get; set; } Gets or sets the path to the ONNX model file.