Question regarding Voxel-based Spatial Pruning implementation

[minrui-liu]

Hi authors,

Thank you for sharing this impressive work.

In streamvln_eval.py, I noticed that historical frame sampling currently uses a fixed-stride strategy:
if self.num_history is None: history_ids = slice(0, time_ids[0], self.num_future_steps) else: history_ids = slice(0, time_ids[0], (time_ids[0] // self.num_history))

However, I could not find an implementation of the voxel-based spatial pruning strategy described in the paper’s Slow-Updating Memory Context, nor other depth-aware processing code.

If possible, Would you mind please share it for me? Many thanks for your time.