f1's placing, re-based: the chain of its segment frames and every quantity that places a part, each in the frame of the segment it
belongs to. This table is the one source of where f1's parts stand; AutoMovieHumanF1Placement maps it through the chain to the
figure's frame at a pose, and AutoMovieHumanF1Joints, AutoMovieHumanF1Landmarks and every family's placing fields read it at the
reference pose.
Generated data, never edited by hand: test/src/reference-conics/FamilyConicRebase.ts (from test/: pnpm conic:build, then
node .shots/esbuild/FamilyConicRebase.cjs --tables) wrote it from the world tables of step 1 (2ad771f825): the chain is
AutoMovieHumanSegmentFrame.chain of the frames AutoMovieHumanF1Frames.segments builds by its written rules, a point is
A^T (p - o) and a direction A^T d in its segment's reference frame. Which segment each belongs to is declared, with its basis,
in that tool (frameOf, LEGS, TRUNK); a note says how a relief's window lies on its course when it crosses a joint.
Every number is written at full double precision, JavaScript's shortest form that reads back the same double, because the round
trip to the reference pose must be exact within rounding and a shortened number would add its own error: the tool enters a value
only when it returns through the chain within 1e-12 mm of its world value for a point and 1e-14 for a direction, and its
receipt lists each return. The reference pose is the chain's rotations and nothing else; identity rotations give the neutral body.
A component under 1e-9 mm (a point, a chain offset) or 1e-12 (a direction, a rotation entry) is written as 0 and nothing else is
rounded: a number that is zero by its definition, such as a joint centre in the segment whose origin it is, leaves the formula as
rounding noise, and step 3 reads its laws here. The thresholds stand far below any datum's resolution (0.01 mm; 1e-5 for a
relief's axis) and far above the double rounding of a 2,000 mm figure (about 4e-13 mm), and a zeroed entry is accepted only
while it still returns within the bounds above.
Millimetres; directions are as their sources wrote them (a relief's axis is rounded to 1e-5, so it is unit only to that).
f1's placing, re-based: the chain of its segment frames and every quantity that places a part, each in the frame of the segment it belongs to. This table is the one source of where f1's parts stand;
AutoMovieHumanF1Placementmaps it through the chain to the figure's frame at a pose, andAutoMovieHumanF1Joints,AutoMovieHumanF1Landmarksand every family's placing fields read it at the reference pose.Generated data, never edited by hand:
test/src/reference-conics/FamilyConicRebase.ts(fromtest/:pnpm conic:build, thennode .shots/esbuild/FamilyConicRebase.cjs --tables) wrote it from the world tables of step 1 (2ad771f825): the chain isAutoMovieHumanSegmentFrame.chainof the framesAutoMovieHumanF1Frames.segmentsbuilds by its written rules, a point isA^T (p - o)and a directionA^T din its segment's reference frame. Which segment each belongs to is declared, with its basis, in that tool (frameOf,LEGS,TRUNK); anotesays how a relief's window lies on its course when it crosses a joint.Every number is written at full double precision, JavaScript's shortest form that reads back the same double, because the round trip to the reference pose must be exact within rounding and a shortened number would add its own error: the tool enters a value only when it returns through the chain within 1e-12 mm of its world value for a point and 1e-14 for a direction, and its receipt lists each return. The reference pose is the chain's rotations and nothing else; identity rotations give the neutral body.
A component under 1e-9 mm (a point, a chain offset) or 1e-12 (a direction, a rotation entry) is written as 0 and nothing else is rounded: a number that is zero by its definition, such as a joint centre in the segment whose origin it is, leaves the formula as rounding noise, and step 3 reads its laws here. The thresholds stand far below any datum's resolution (0.01 mm; 1e-5 for a relief's axis) and far above the double rounding of a 2,000 mm figure (about 4e-13 mm), and a zeroed entry is accepted only while it still returns within the bounds above.
Millimetres; directions are as their sources wrote them (a relief's
axisis rounded to 1e-5, so it is unit only to that).