[main] Downsample transcripts for quadtree tiling construction - #83
Open
Tobiaspk wants to merge 3 commits into
Open
[main] Downsample transcripts for quadtree tiling construction#83Tobiaspk wants to merge 3 commits into
Tobiaspk wants to merge 3 commits into
Conversation
Subsamples points before building the training-tile quadtree once the count exceeds max_transcripts (default 50M), scaling max_size and its retry step by the same fraction. Bounds/scale/max_depth still use the full point set, so leaf boundaries stay comparable to the full-dataset tree.
Tobiaspk
changed the title
Collaborator
Author
|
Almost Identical tiling results as compared to the full 1B quadtree tiling tree. 
Performance not improved, but this avoids numeric overflows: 
|
Collaborator
Author
|
Defaults max_transcripts to 50 Mio. |
…h construction The tiling_max_transcripts CLI default (50M) only reached QuadTreeTiling; points_in_polygons/get_quadtree_index calls made during prediction-graph construction never received it, so quadtree overflows (issue dpeerlab#77) weren't guarded on that path.
max_transcripts read as a hard cap on transcripts; the parameter only downsamples points fed into quadtree construction, so the old name was misleading throughout the CLI, ISTDataModule, and geometry/tiling call chain.
Collaborator
Author
|
Will probably deprecate this in favor of the upcoming |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #77: adds an optional
max_transcriptsparameter (default 50M) that builds the training-tile quadtree on a random subsample once the point count exceeds it, scalingmax_sizeby the same fraction so leaf density still approximates the full-dataset tree.This limits GPU memory and avoids cuSpatial's numeric overflow issues (especially in atera), and improving quadtree tiling stability. it does not improve construction time, since the bottleneck is computing global bounds/scale over all points.