Height-aware point sampling for forest semantic segmentation
- Label
- First paper · height sampling
- Venue
- Undecided (IEEE GRSL letter or CVPR EarthVision workshop)
- Status
- planning
A first, learning-oriented paper. It measures how point-selection rules at a fixed point budget change understory segmentation, with height-stratified selection as the reference that later methods must beat.
Experiments
- Notebook 01: PointNet and PointNet++ baselines on ForestSemantic-MS
- Notebook 02: sampling retention gate
Tasks
- 16 Oct 2026 · NB01: PointNet and PointNet++ baselines · Done
- 16 Oct 2026 · NB02: sampling retention gate · Done
- 30 Oct 2026 · Rerun the retention gate on sparser UAV data (FOR-instance) · Planned
- 13 Nov 2026 · Class-weighted loss as a rebalancing control · Planned