Comparisons between feature extraction and matching algorithms. Learning-based feature extraction and matching algorithms like SuperPoint and SuperGlue outperform traditional methods such as SIFT and NN.
![](https://mscvprojects.ri.cmu.edu/2024team4/wp-content/uploads/sites/102/2024/05/iShot_2024-05-04_23.03.18-1024x108.png)
![](https://mscvprojects.ri.cmu.edu/2024team4/wp-content/uploads/sites/102/2024/05/iShot_2024-05-04_23.03.25.png)
![](https://mscvprojects.ri.cmu.edu/2024team4/wp-content/uploads/sites/102/2024/05/iShot_2024-05-04_23.04.45-1024x619.png)
- Both qualitative and quantitative experiments demonstrate the superior performance of learning-based methods, such as SuperPoint and SuperGlue (SP&SG), compared to traditional approaches like SIFT, adalma, and NN.
- Besides the number of keypoints and matches, SP&SG is faster than SIFT&NN.
![](https://mscvprojects.ri.cmu.edu/2024team4/wp-content/uploads/sites/102/2024/05/iShot_2024-05-04_23.06.55-1024x457.png)
Future Work
We plan to leverage the new end-to-end deep-learning based method VGGSfM which is fully differentiable to get subpixel accuracy and precise intrinsics on the Multiface dataset.