Open CvierXi opened 2 years ago
Thank you for your reminder. The bug has been fixed https://github.com/openxrlab/xrlocalization/commit/36f6fbe61e5f5c58ed20eaf16498ebcecb5dae91
We recommend you use SuperPoint, which can provide faster and more accurate localization in common scenes. The usage can be referred to here.
Hi,thank you for your answer.A new problem occurred when I used the fix file in xrlocalization.Can you give me a hand ?Thank you. Here is what went wrong. Step 3: Re-extract/match with superpoint 100%|█████████████████████████████████████████| 540/540 [00:58<00:00, 9.25it/s]
100%|██████████████████████████████████████| 7162/7162 [00:52<00:00, 136.84it/s] Step 4: Re-triangulate terminate called after throwing an instance of 'std::bad_alloc' what(): std::bad_alloc recon.sh:行 56: 8591 已放弃 (核心已转储) ./bin/run_triangulation {SFM_DIR}/features.bin {REFINE_DIR} run triangulation failed
https://github.com/openxrlab/xrlocalization/blob/b0756b220d9f425379ce6ab4e529c5a3d24002ed/xrloc/features/d2net.py#L50
D2Net here output features with 512 dim. But feature I/O and XRSfm only support 256 dim by default.