feat: navigable 3D embedding view - #217
Draft
somehowchris wants to merge 2 commits into
Draft
Conversation
Add end-to-end 3D support: points can carry a Z coordinate and render as a navigable 3D point cloud with orbit/pan/zoom, fog depth-cueing, and exact 3D picking, on both the WebGPU and WebGL2 backends. Component - camera3d: orbit/pan/dolly camera with fit-to-bounds, plus frustum-aware downsampling (chunked, idle-scheduled, cancelable) bounded by downsampleMaxPoints so multi-million-point clouds stay responsive. - WebGPU + WebGL2 renderers: 3D billboard draw, depth-tested pick targets (r32uint / RGBA8), and fog. - EmbeddingViewImpl: orbit inertia, reset/focus tweens, 3D hover/selection, a status-bar reset-camera control, and a visible active-filter indicator while the brush outline is hidden in 3D (never a silent hidden filter). - EmbeddingViewMosaic: exact 3D pick-by-index resolution — rowid for base tables, the configured identifier (size-gated to stay bounded) for rowid-less views/joins, coordinate fallback only when no exact identity exists. Coordinated-selection predicates prefer the semantic identifier, then rowid, then coordinates, so a pick targets exactly the picked row. Viewer - A "Z (optional)" projection-column selector; choosing a Z column opens a navigable 3D view that round-trips through saved charts. Backend - --z to pass a precomputed Z column, and --umap-n-components 3 to generate a 3D projection. Tests - camera3d, matrix, and mosaic_client (pick identity + predicate ordering) unit tests; backend CLI and projection coverage. Wires the component package into the JS test suite. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Member
|
Could you share a few screenshots and maybe a video? Would love to see this even though we obviously can't take on such a big maintenance commitment. |
Follow-up pass on the navigable 3D view: fixes real robustness/pick bugs, removes code that did not earn its place, and closes 2D/3D coherence gaps. Correctness - Guard pointsBounds3D against NaN/Inf coordinates so a null/NaN z can no longer poison the centroid into a blank-view camera. - Stop a focus double-click from first clearing a hidden 3D filter; in 3D the filter now clears only on an empty-space click. - Guard the rowid probe against a shadowing user column named `rowid`. - Cast SUM/MAX(count) to BIGINT (overflow) and convert to Number at the JS boundary so density init never throws on bigint arithmetic. Cleanup - Delete the production-dead frustumSampleIndices/emitFrustumSample and matrix4_transform_point; fix the stale "Web Worker" comment. - Remove the size-gated identifier-materialization pick tier; pick identity is now rowid -> coordinate+z. The coordinate fallback returns indeterminate when distinct identifiers collide at the exact rendered coordinates (COUNT DISTINCT), rather than selecting an arbitrary sibling. - Refactor EmbeddingViewImpl to accept x/y/z/category as separate reactive props, dissolving ~10 shadow variables and the data-wrapper churn machinery. 3D UX parity - Scroll zoom-to-cursor, density-adaptive point size, occlusion-aware tooltips, perspective-aware highlight ring. Coherence & backend - Expose and persist fogDensity in the viewer; export Camera3DState from the umbrella package; fix the stale points-3d Select value. - Error on --umap-n-components 3 with precomputed --x/--y or --disable-projection; validate the --z column exists and is numeric. Tests & docs - 61 component tests (queryClosestPoint3D, rowid/BIGINT paths, bounds NaN, degenerate matrix) and backend CLI coverage; README/docs mention 3D. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
2 participants
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.
FYI, this has been fully vibe coded over a Sunday afternoon and only tested, only validated with own datasets.
AI generated overview:
Add end-to-end 3D support: points can carry a Z coordinate and render as a navigable 3D point cloud with orbit/pan/zoom, fog depth-cueing, and exact 3D picking, on both the WebGPU and WebGL2 backends.
Component
Viewer
Backend
Tests