notes.txt

### Notes

- In mapper.ini keep transitivity high.
- Processes in parrallel. Extraction of featues (each image a separate process). Also matching cam be done in parallel.
- Visualize in real time optical flow for keyframes (is a different local feature)
- To save time, save local features from keyframe selection
- For mapper, use tie points only from last 6 frames
- pass Adalam to GPU, it is enough to import config file (https://kornia.readthedocs.io/en/latest/_modules/kornia/feature/adalam/adalam.html#match_adalam)
- Check adalam and lightglue, LAF correctly implemented. See https://github.com/ducha-aiki/imc2023-kornia-starter-pack/blob/main/DISK-adalam-pycolmap-3dreconstruction.ipynb
            kpoints, des, images = import_local_features.ImportLocalFeature(
                cfg.DATABASE ######################################################################### Non ha senso importare keypoints, descriptors che poi vengono sovrascritti Fatto per descrittori > 128
            )
            true_indices = np.where(adjacency_matrix)
For performance optimization:
- [ ] Do not save txt reconstruction (only binary)

(
                cfg.DATABASE ######################################################################### Non ha senso importare keypoints, descriptors che poi vengono sovrascritti Fatto per descrittori > 128
            )

### CALIBRATIONS
# Machine Hall
N_CAMERAS = 2
CAM0 = 4,752,480,458.654,457.296,367.215,248.375,-0.28340811,0.07395907,0.00019359,1.76187114e-05
CAM1 = 4,752,480,457.587,456.134,379.999,255.238,-0.28368365,0.07451284,-0.00010473,-3.55590700e-05
BASELINE_CAM0_CAM1 = 0.110078

# GoPro webcam (Wide)
N_CAMERAS = 1
CAM0 = 5,1920,1080,992.12798,992.12798,540,495,0.0354746,-0.0039797,-0.000946751,-0.0127286