How jirrrr77/video_sychronazation works
This thesis repository experiments with synchronizing frames from two video perspectives. The documented workflow tries two alternative matching approaches, applies a person-focused background mask, then evaluates alignment with Sampson distance. The README says these experiments did not work reliably and that the author moved toward optical flow; no optical-flow implementation is visible in the supplied tree. Notebook internals were not sampled, so links below reflect documented composition rather than verified function calls.
Main entry points: loFTR_mrcnn_sampson.ipynb LoFTR experiment · denseMatching_mrcnn_sampson.ipynb Dense matching experiment · background_mask.py Person background mask · utils.py Sampson distance
Components 7
Matching Experiments
- LoFTR experiment
code/loFTR_mrcnn_sampson.ipynb - Dense matching experiment
code/denseMatching_mrcnn_sampson.ipynb
Frame Preprocessing
- Person background mask
code/background_mask.py
Geometry Utilities
- Sampson distance
code/utils/utils.py
Other components
- Researcher
- Two-view frames
- Camera calibration
data/calibration_150821_dance5.json
Connections 7
- Researcher → Two-view frames: provides
- Two-view frames → LoFTR experiment: aligns frames
- Two-view frames → Dense matching experiment: aligns frames
- LoFTR experiment → Person background mask: uses masking
- LoFTR experiment → Sampson distance: evaluates matches
- Dense matching experiment → Person background mask: uses masking
- Dense matching experiment → Sampson distance: evaluates matches