Crate loftr 

Native Rust/tch LoFTR feature matching.

This crate exposes a small, high-level API for constructing a LoFTR model, loading compatible weights, and matching a pair of images.

Typical usage:

1. Build a LoftrModel with a preset LoftrConfig.
2. Load exported weights with LoftrModel::load_weights.
3. Run LoftrModel::forward or LoftrModel::forward_debug.

The published docs are built with the doc-only feature so they can render on docs.rs without linking libtorch. For local runs, either point tch at an existing libtorch install or enable the download-libtorch feature.

Example

use loftr::{LoftrConfig, LoftrModel};
use tch::{Device, Kind, Tensor};

let mut model = LoftrModel::new(Device::Cpu, LoftrConfig::outdoor())?;
model.load_weights("artifacts/weights/loftr_outdoor_state_dict.safetensors")?;

let image0 = Tensor::rand([1, 1, 128, 128], (Kind::Float, Device::Cpu));
let image1 = Tensor::rand([1, 1, 128, 128], (Kind::Float, Device::Cpu));
let matches = model.forward(&image0, &image1)?;

println!("match count = {}", matches.confidence.size()[0]);

Structs

CoarseDebugStats

Summary statistics for coarse-stage matching outputs.

FineConfig

Transformer parameters for the fine matching refinement stage.

LoftrConfig

High-level LoFTR model configuration.

LoftrDebugStages

Debug-stage summary returned by LoFTRModel::forward_debug.

LoftrMatches

Matched keypoints and confidences returned by LoFTR inference.

LoftrModel

LoFTR model instance with owned weights and inference state.

MatchCoarseConfig

Matching parameters for the coarse LoFTR correspondence stage.

ResNetFpnConfig

Backbone parameters for the ResNetFPN encoder used by LoFTR.

TensorDebugStats

Summary statistics for a tensor captured during debug inference.

TransformerConfig

Transformer parameters shared by LoFTR coarse and fine stages.

Enums

AttentionType

Attention implementations supported by the LoFTR transformers.

BackboneType

Backbone variants supported by this crate.

LoftrError

Errors returned by the public loftr API.

MatchType

Coarse matcher implementations supported by this crate.

TransformerLayer

Transformer layer ordering used by LoFTR encoder stacks.

Functions

normalize_loftr_image

Converts supported LoFTR image layouts into normalized [B, 1, H, W] tensors.