This summarises how far along the project is. The markings [ ], [~], and [x] denote an incomplete, partially complete, and completed task, respectively.
- Data loading:
- KITTI data tools and prep
- Data augmentations for training:
- Perturbation
- Global scaling
- Global rotation
- [~] Feature learning network:
- Partitioning points
- Grouping points into voxelgrid
- Sampling points within voxels
- [~] (Sparse tensor representation using hash)
- Augmenting voxel-cloud with offsets
- VFE-n custom layers
- Fully-connected component
- Elementwise maxpool
- Pointwise concatenation
- VFE-n custom layer
- VFE-out custom layer
- Convolutional middle layers:
- Convolutional middle block custom layer
- Convolutional middle layers
- Region proposal network:
- RPN Convolutional Block custom layer
- RPN network
- [~] Loss function:
- SmoothL1 loss function
- Loss anchors
- Loss custom layer
- Training
- Optimizations
- [~] Sparse tensor representation
- Sparse operations over tensors
- [~] Fit CARLA lidar to KITTI specifications
- [~] Integrate into CARLA