DeepToPoints

Resolve 21 marks the Deep Image Compositing Toolset as Studio Version Only. A Deep Image may hold multiple samples at different depths inside one screen pixel; DeepToPoints turns those samples into viewable 3D points.

The data boundary matters: the orange Input expects a true Deep Image, while the output is classic Fusion scene3d. This is not a flat RGBA+Z conversion and it is not a file-based point-cloud importer; PointCloud3D serves that separate role.

Camera is a dedicated green input. Blackmagic states that connecting the scene camera accounts for its perspective and precisely positions the point cloud relative to the original 3D scene. The Deep Input remains the only required connection; Camera is optional but important for camera-matched spatial placement.

Style switches between cross-hair and point display, Size changes marker size, and Antialiasing softens point edges. The manual describes Density as controlling the overall concentration/intensity of points in the viewer; do not claim that it synthesizes new geometry or new Deep samples.

Use Per-Point Colors takes color from the connected source. When disabled, Default Color is used instead. This changes the point-cloud representation, not the upstream Deep Image data.

Make Renderable separates a viewer/reference cloud from intentional camera output. When it is enabled, Unseen by Camera appears and can keep the point cloud visible in 3D viewers while excluding it from Renderer3D camera output.

Scale changes the overall cloud size, Depth Scale expands or contracts its depth extent, and Flip Depth inverts the depth data. These controls solve different problems and should not be used interchangeably.

The Transform tab moves the resulting 3D object with Translation, Rotation, Pivot, Scale, Use Target and Import Transform. Import Transform imports transform data only from .lws, .ase or .ma; it is not a geometry/light/camera importer.

DeepToPoints is best treated as a diagnostic and integration bridge: inspect Deep samples in Z, verify camera/depth conventions, and insert the point representation into a normal Merge3D/Renderer3D scene. A discrete point cloud is not automatically a reconstructed continuous mesh surface.

**Do not confuse the two Camera interfaces.** The green Camera socket in Flow accepts a 3D camera as a separate connection. The original Controls capture also shows a Camera dropdown set to Default. Do not treat that dropdown as a substitute for the physical socket; the exact conditional behavior needs a live Resolve check.

**The Controls screenshot is evidence, not a universal preset.** The source capture shows Size 1, Density 0.1, Scale 0.005, Depth Scale 0.025 and Flip Depth enabled. These values are not suitable for every Deep EXR: verify the depth convention, source camera and scene scale first.

**Viewer display does not imply rendered output.** A cloud may be useful in the 3D Viewer for checking near and far samples while Make Renderable is off. To get markers into MediaOut you need Renderer3D and camera-visible rendering enabled.

**Unseen by Camera is conditional, not visible in the saved capture.** The Resolve 21 manual says the checkbox appears when Make Renderable is enabled. The saved screenshot has Make Renderable off, so the checkbox is absent; that does not demonstrate that the feature was removed.

**Cross and Point are marker display styles.** Cross makes nearby samples easier to distinguish when navigating the cloud, while Point can be more compact. Switching styles does not produce a polygon mesh or modify the source sample structure.

**Density changes display concentration.** The manual describes overall concentration and visual intensity of displayed points. Do not claim that increasing Density recovers missing geometry or adds source Deep samples; compare values against a fixed input.

**Per-point colors and Default Color serve alternative display modes.** With Use Per-Point Colors enabled, point colors come from the source clip. When disabled, the Default Color swatch and its visible RGB channels provide a uniform marker color for reading spatial structure.

**Prove camera matching before compensating with scale.** When cloud points diverge from CG geometry, check the camera model, frame, FOV and pose first. Global Scale or Flip Depth may conceal a mismatch in one region without restoring the correct projection.

**Controls Scale and Transform Scale belong to different stages.** The first sets the general cloud size from Deep data; the second acts on the resulting 3D object alongside Translation, Rotation, Pivot and optional per-axis scaling. Avoid compensating with both without a controlled comparison.

**Test Flip Depth against known near/far ordering.** The manual defines it as an inversion of depth data along Z. Without reference geometry it is hard to distinguish a correct depth convention from merely a different-looking cloud, particularly if Camera and Depth Scale also change.

**Rotation Order means six axis sequences, not six rotation axes.** XYZ, XZY, YXZ, YZX, ZXY and ZYX define how the three rotations are composed. The difference matters for large rotations, so record the order when matching an exported 3D scene.

**Use Target changes orientation behavior.** When enabled, the object aims toward an XYZ target and Rotation becomes target-relative. It is not a camera tracker or a way to automatically reconstruct a cloud from a 2D video frame.

**Import Transform imports transform data only.** Supported formats include .lws, .ase and .ma; this button does not import 3D geometry, lights or cameras. Verify position, rotation and scale after the import before modifying the cloud geometry.

**Validation boundary: diagrams are not renders.** The existing Flow lessons check Deep/scene3d/image types and named ports, but cannot prove correct Deep EXR reconstruction, camera calibration or Renderer3D pixels. Native acceptance still requires Resolve Studio and known reference frames.