Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed
This work addresses efficiency bottlenecks in semi-dense local feature matching for applications like image retrieval and 3D reconstruction, though it is incremental as it builds directly on LoFTR.
The paper tackles the low efficiency of the LoFTR feature matching method by proposing an aggregated attention mechanism and a two-stage correlation layer, resulting in a model that is about 2.5 times faster than LoFTR while achieving higher accuracy.
We present a novel method for efficiently producing semi-dense matches across images. Previous detector-free matcher LoFTR has shown remarkable matching capability in handling large-viewpoint change and texture-poor scenarios but suffers from low efficiency. We revisit its design choices and derive multiple improvements for both efficiency and accuracy. One key observation is that performing the transformer over the entire feature map is redundant due to shared local information, therefore we propose an aggregated attention mechanism with adaptive token selection for efficiency. Furthermore, we find spatial variance exists in LoFTR's fine correlation module, which is adverse to matching accuracy. A novel two-stage correlation layer is proposed to achieve accurate subpixel correspondences for accuracy improvement. Our efficiency optimized model is $\sim 2.5\times$ faster than LoFTR which can even surpass state-of-the-art efficient sparse matching pipeline SuperPoint + LightGlue. Moreover, extensive experiments show that our method can achieve higher accuracy compared with competitive semi-dense matchers, with considerable efficiency benefits. This opens up exciting prospects for large-scale or latency-sensitive applications such as image retrieval and 3D reconstruction. Project page: https://zju3dv.github.io/efficientloftr.