NeuralTracking
NeuralTracking copied to clipboard
Implement classic DynamicFusion t-minus-one-to-t scheme
- [x] Penalty loss functions
- [x] Point-to-plane distances for mesh + point cloud
- [x] Graph --> CPU+CUDA C++ WarpField implementation + test (#17)
- [x] Integrate CUDA+CPU C++ K-D Tree into WarpField #29
- [x] Minor redesign of Warpable Voxel Grid (to be acted on by WarpField, somehow?)
- [x] Data Term Residuals Skeleton
- [x] Data term residuals -- normals estimation (CPU/CUDA)
- [x] Data term residuals -- normal projection using rendering
- [x] Data term -- Jacobian computation
- [x] ~Rigid ICP for extrinsic matrix estimation~ Outsourced to #46
- [x] Warping mesh with extrinsic matrix
- [x] Render-based Data Term Residuals
- [x] Data Term Energy
- [x] Hierarchical Graph Data Structure
- [x] Hierarchical Graph Resampling via Knn / K-D Tree #32
- [x] ~Node updates~ Outsourced to 48
- [x] Regularization Term Residuals
- [x] Regularization Term Energy
- [x] Apply linear solver + conditioning number check (?)
- [x] ~Test on 2 frames~ -- will handled as separate issue
Full math worked out for computing the data term Jacobian here.