GS::MASt3R::Model
Full MASt3R model for dense stereo
Constructors
new(config : ModelConfig, device : Tensor::Device = Tensor::Device::GPU)
SourceInstance methods
config
Sourcedecoder
Sourceencoder
Sourceforward_pair(img1 : Autograd::Variable, img2 : Autograd::Variable) : Tuple(Autograd::Variable, Autograd::Variable, Autograd::Variable, Autograd::Variable)
Process image pair with cross-attention (full MASt3R) img1, img2: [batch, channels, height, width] Returns: {xyz1, conf1, xyz2, conf2}
Process a single image (no cross-attention) img: [batch, channels, height, width] Returns: {xyz: [batch, h, w, 3], conf: [batch, h, w, 1]}
parameters
Sourcepointmap_head
Source