IconVSROptions
Options & Configuration · AiDotNet.Video.Options
Configuration options for IconVSR, BasicVSR with information refill and coupled propagation.
Inheritance: ModelOptions → NeuralNetworkOptions → IconVSROptions
For Beginners
In long videos, processing frames one-by-one can accumulate errors. IconVSR picks “keyframes” as reliable reference points, re-injecting features computed directly from the neighboring frames there, and lets the forward pass see what the backward pass found.
How It Works
Defaults follow Chan et al., CVPR 2021, and the authors’ release (open-mmlab/mmagic, IconVSRNet and configs/iconvsr/iconvsr_2xb4_reds4.py): 64-channel features, 30 residual blocks per propagation branch, keyframes every 5 frames refilled from an EDVR-M feature extractor over 5 frames (64 channels, 8 deformable groups, 5 extraction blocks), x4 upsampling, 15-frame training clips, Charbonnier loss, and BasicVSR’s schedule: Adam (0.9, 0.99) at 2e-4 with SPyNet at 0.125 times the rate, SPyNet and EDVR frozen for the first 5,000 iterations, cosine-annealed to 1e-7 over 300,000 iterations.
Constructors
| Constructor | Summary |
|---|---|
IconVSROptions() |
Initializes a new instance with default values. |
IconVSROptions(IconVSROptions 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 CharbonnierEpsilon { get; set; } |
Gets or sets the Charbonnier ε; the release’s loss is sqrt(d² + 1e-12), so ε = 1e-6. |
double LearningRate { get; set; } |
Gets or sets Adam’s starting learning rate (2e-4). |
double MinLearningRate { get; set; } |
Gets or sets the floor of the cosine schedule (1e-7). |
double SpyNetLearningRateMultiplier { get; set; } |
Gets or sets the multiplier of SPyNet’s learning rate (0.125, i.e. |
int CosineIterations { get; set; } |
Gets or sets the iterations of the cosine schedule (300,000). |
int DeformableGroups { get; set; } |
Gets or sets the deformable groups of EDVR’s PCD alignment (8). |
int EdvrChannels { get; set; } |
Gets or sets the EDVR-M feature extractor’s channels (64). |
int EdvrExtractionBlocks { get; set; } |
Gets or sets the EDVR extractor’s per-frame residual blocks (5). |
int FlowFreezeIterations { get; set; } |
Gets or sets the iterations SPyNet and the EDVR extractor stay frozen at the start of training (5,000). |
int KeyframeStride { get; set; } |
Gets or sets the keyframe spacing: frames 0, s, 2s, … |
int MidChannels { get; set; } |
Gets or sets the feature channels of the propagation branches (64). |
int NumFrames { get; set; } |
Gets or sets the frames of a training clip (15, the REDS recipe). |
int NumResBlocks { get; set; } |
Gets or sets the residual blocks per propagation branch (30). |
int RefillPadding { get; set; } |
Gets or sets the frames on each side of a keyframe the refill extractor sees (2 for REDS, 3 for Vimeo-90K). |
string ModelPath { get; set; } |
Gets or sets the path to the ONNX model file. |