GS::MASt3R::MASt3RInference
Full MASt3R model for inference
Constructors
new(device : Tensor::Device = Tensor::Device::CPU)
SourceInstance methods
decoder
Sourcedevice
Sourcedpt_head1
Sourcedpt_head2
Sourceencoder
SourceInference on image pair img1, img2: Tensor [batch, height, width, 3] in range [0, 1] Returns: {points1, conf1, points2, conf2} points: [batch, H4, W4, 3] - 3D coordinates (upsampled) conf: [batch, H4, W4, 1] - confidence
Inference on single image (no cross-attention)
frames_to_pointcloud(frames : Array(Video::Frame), conf_threshold : Float32 = 0.3_f32) : Tuple(Tensor, Tensor)
Convert frames to point cloud frames: Array of Tensor [height, width, 3] Returns: points [N, 3], confidences [N]
weights_loaded?
Source