LoFTR: Detector-Free Local Feature Matching with Transformers
This addresses the challenge of feature matching in low-texture areas for computer vision applications, offering a novel approach that improves performance over existing methods.
The paper tackles the problem of local image feature matching by proposing a detector-free method that uses Transformers to establish dense matches at a coarse level and refine them, outperforming state-of-the-art methods by a large margin on indoor and outdoor datasets and ranking first on two visual localization benchmarks.
We present a novel method for local image feature matching. Instead of performing image feature detection, description, and matching sequentially, we propose to first establish pixel-wise dense matches at a coarse level and later refine the good matches at a fine level. In contrast to dense methods that use a cost volume to search correspondences, we use self and cross attention layers in Transformer to obtain feature descriptors that are conditioned on both images. The global receptive field provided by Transformer enables our method to produce dense matches in low-texture areas, where feature detectors usually struggle to produce repeatable interest points. The experiments on indoor and outdoor datasets show that LoFTR outperforms state-of-the-art methods by a large margin. LoFTR also ranks first on two public benchmarks of visual localization among the published methods.