PXD - Revision history

AdelQue at 05:05, 3 March 2026

2026-03-03T05:05:38Z

← Older revision		Revision as of 05:05, 3 March 2026
Line 1:		Line 1:
	{{Notice\|type=warn\|content=This page is unfinished.}}		{{Notice\|type=warn\|content=This page is unfinished.}}

			===== If you are looking for tools to edit <code>.pxd</code> files, see [[PXD Animation Tools\|PXD Animation Tools.]] =====

	== PXD Files ==		== PXD Files ==

AdelQue: Corrected Olympic game

2025-04-16T23:55:48Z

Corrected Olympic game

← Older revision		Revision as of 23:55, 16 April 2025
Line 2:		Line 2:

	== PXD Files ==		== PXD Files ==
	Hedgehog Engine 2 games from [[Mario & Sonic at the ~~Rio 2016~~ Olympic Games\|~~Rio 2016 Olympics~~]] to [[Shadow Generations]] contain skeleton and skeletal animation files with the file extensions <code>.skl.pxd</code> and <code>.anm.pxd</code> respectively, stored in a [[BINA]] container. Skeletons contain pose data that usually result in a T-Pose, bone and parent indices, and bone names. Animations contain playback metadata such as frame rate and frame count, track count, and either compressed animation pose data using [https://github.com/nfrechette/acl ACL compression,] or raw uncompressed pose data that use bone and frame indices to specify keyframed transforms, with linear interpolation between keyframes. For both skeletons and animations, position and rotation data for each bone is an absolute transform relative to the parent position, whereas scales are inherited from their respective parents locally, and have no effect on their positions.		Hedgehog Engine 2 games from [[Mario & Sonic at the Tokyo 2020 Olympic Games\|Tokyo 2020 Olympic Games]] to [[Shadow Generations]] contain skeleton and skeletal animation files with the file extensions <code>.skl.pxd</code> and <code>.anm.pxd</code> respectively, stored in a [[BINA]] container. Skeletons contain pose data that usually result in a T-Pose, bone and parent indices, and bone names. Animations contain playback metadata such as frame rate and frame count, track count, and either compressed animation pose data using [https://github.com/nfrechette/acl ACL compression,] or raw uncompressed pose data that use bone and frame indices to specify keyframed transforms, with linear interpolation between keyframes. For both skeletons and animations, position and rotation data for each bone is an absolute transform relative to the parent position, whereas scales are inherited from their respective parents locally, and have no effect on their positions.

	== <code>.anm.pxd</code> Structure ==		== <code>.anm.pxd</code> Structure ==
Line 58:		Line 58:
	=== Native Uncompressed Data: ===		=== Native Uncompressed Data: ===
	{{Notice\|type=note\|content=This section could use a less confusing explanation.}}		{{Notice\|type=note\|content=This section could use a less confusing explanation.}}
	Alternatively to compressed ACL track data, the PXAN structure can store individual components (rotation, translation, scale) of any given track at any frame, and linearly interpolate between them independently. This system was first seen in the [[Mario & Sonic at the Rio 2016 Olympic Games\|M&S ~~2016 Rio~~ Olympic Games]] as the exclusive method of storing animations. ACL compression may have an advantage in reducing the file size of many of these animations but—though rare—uncompressed animations have been observed in subsequent games for certain animations, presumably to combat any potential jitter during close up cutscenes or similar situations. Though the actual end data contained in this method is simple, the structure can end up like a spider's nest at first glance:		Alternatively to compressed ACL track data, the PXAN structure can store individual components (rotation, translation, scale) of any given track at any frame, and linearly interpolate between them independently. This system was first seen in the [[Mario & Sonic at the Rio 2016 Olympic Games\|M&S Tokyo 2020 Olympic Games]] as the exclusive method of storing animations. ACL compression may have an advantage in reducing the file size of many of these animations but—though rare—uncompressed animations have been observed in subsequent games for certain animations, presumably to combat any potential jitter during close up cutscenes or similar situations. Though the actual end data contained in this method is simple, the structure can end up like a spider's nest at first glance:

	==== Track Offset Tables: ====		==== Track Offset Tables: ====

AdelQue at 09:14, 8 April 2025

2025-04-08T09:14:01Z

← Older revision		Revision as of 09:14, 8 April 2025
Line 78:		Line 78:

	==== Frame Tables and Transforms: ====		==== Frame Tables and Transforms: ====
	Each transform type will get a frame table as <code>uint16_t[key_count] keys;</code> where each value is the frame that the keyframe will be inserted to, and transform array <code>vector4f[key_count] transforms; // XYZW floats</code> that will correspond to the frame with the matching array index. Unlike in ACL decompressed tracks, native uncompressed tracks' <code>W</code> component is always <code>0.0</code>. Both of these element arrays will be null aligned to 16 bytes.		Each transform type will get a frame table as <code>uint16_t[key_count] keys;</code> where each value is the frame that the keyframe will be inserted to, and transform array <code>vector4f[key_count] transforms; // XYZW floats</code> that will correspond to the frame with the matching array index. Unlike in ACL decompressed tracks, native uncompressed tracks' <code>W</code> component for translation and scale is always <code>0.0</code>. Both of these element arrays will be null aligned to 16 bytes.

	==== Posing and Transform Inheritance: ====		==== Posing and Transform Inheritance: ====

AdelQue at 08:57, 8 April 2025

2025-04-08T08:57:01Z

← Older revision		Revision as of 08:57, 8 April 2025
Line 104:		Line 104:
	* [https://github.com/ik-01 ik-01]: Format and game code research, filling in gaps for unknown values		* [https://github.com/ik-01 ik-01]: Format and game code research, filling in gaps for unknown values
	* [https://github.com/AdelQue AdelQue:] Format research, math relations and bone inheritance behaviors		* [https://github.com/AdelQue AdelQue:] Format research, math relations and bone inheritance behaviors
			[[Category:File Formats]]

AdelQue at 08:42, 8 April 2025

2025-04-08T08:42:22Z

← Older revision		Revision as of 08:42, 8 April 2025
Line 80:		Line 80:
	Each transform type will get a frame table as <code>uint16_t[key_count] keys;</code> where each value is the frame that the keyframe will be inserted to, and transform array <code>vector4f[key_count] transforms; // XYZW floats</code> that will correspond to the frame with the matching array index. Unlike in ACL decompressed tracks, native uncompressed tracks' <code>W</code> component is always <code>0.0</code>. Both of these element arrays will be null aligned to 16 bytes.		Each transform type will get a frame table as <code>uint16_t[key_count] keys;</code> where each value is the frame that the keyframe will be inserted to, and transform array <code>vector4f[key_count] transforms; // XYZW floats</code> that will correspond to the frame with the matching array index. Unlike in ACL decompressed tracks, native uncompressed tracks' <code>W</code> component is always <code>0.0</code>. Both of these element arrays will be null aligned to 16 bytes.

	Posing and Transform Inheritance:		==== Posing and Transform Inheritance: ====

	Each bone's transformation is stored as an object space (AKA 'Pose Space' or 'Model Space') transform relative to its parent bone's final transformation. That means, regardless of what the reset pose or bone orientations are in the model's skeleton, the stored values of any given frame will completely overwrite any pose previously present.		Each bone's transformation is stored as an object space (AKA 'Pose Space' or 'Model Space') transform relative to its parent bone's final transformation. That means, regardless of what the reset pose or bone orientations are in the model's skeleton, the stored values of any given frame will completely overwrite any pose previously present.

Line 92:		Line 91:
	When reading in-game, though scales from recursive parents are inherited and multiplied, translation is completely unaffected in any capacity, unintuitively. Therefore, before applying any object space transformation in a setting where parent scale is assumed to affect the child bone's location, each bone's final scale must be calculated and applied by multiplying its and each recursive parent's scales together.		When reading in-game, though scales from recursive parents are inherited and multiplied, translation is completely unaffected in any capacity, unintuitively. Therefore, before applying any object space transformation in a setting where parent scale is assumed to affect the child bone's location, each bone's final scale must be calculated and applied by multiplying its and each recursive parent's scales together.

	Trivia: The inheritance and relationships behavior for bone transform values ~~are~~ identical to all previous HE games that used Havok, possibly indicating that Sonic Team's custom implementation of animation playback and posing is loosely based on Havok's <code>~~hkaPose~~</code>.		'''Trivia''': ''The inheritance and relationships behavior for bone transform values have been found to be identical to all previous HE games that used Havok, possibly indicating that Sonic Team's custom implementation of animation playback and posing is loosely based on Havok's <code>hka</code> functions.''

	== <code>.skl.pxd</code> Structure ==		== <code>.skl.pxd</code> Structure ==

AdelQue: Add Thanks

2025-04-08T08:40:39Z

Add Thanks

← Older revision		Revision as of 08:40, 8 April 2025
Line 58:		Line 58:
	=== Native Uncompressed Data: ===		=== Native Uncompressed Data: ===
	{{Notice\|type=note\|content=This section could use a less confusing explanation.}}		{{Notice\|type=note\|content=This section could use a less confusing explanation.}}
	~~<sub>~~Alternatively to compressed ACL track data, the PXAN structure can store individual components (rotation, translation, scale) of any given track at any frame, and linearly interpolate between them independently. This system was first seen in the [[Mario & Sonic at the Rio 2016 Olympic Games\|M&S 2016 Rio Olympic Games]] as the exclusive method of storing animations. ACL compression may have an advantage in reducing the file size of many of these animations but—though rare—uncompressed animations have been observed in subsequent games for certain animations, presumably to combat any potential jitter during close up cutscenes or similar situations. Though the actual end data contained in this method is simple, the structure can end up like a spider's nest at first glance:~~</sub>~~		Alternatively to compressed ACL track data, the PXAN structure can store individual components (rotation, translation, scale) of any given track at any frame, and linearly interpolate between them independently. This system was first seen in the [[Mario & Sonic at the Rio 2016 Olympic Games\|M&S 2016 Rio Olympic Games]] as the exclusive method of storing animations. ACL compression may have an advantage in reducing the file size of many of these animations but—though rare—uncompressed animations have been observed in subsequent games for certain animations, presumably to combat any potential jitter during close up cutscenes or similar situations. Though the actual end data contained in this method is simple, the structure can end up like a spider's nest at first glance:

	==== Track Offset Tables: ====		==== Track Offset Tables: ====
Line 80:		Line 80:
	Each transform type will get a frame table as <code>uint16_t[key_count] keys;</code> where each value is the frame that the keyframe will be inserted to, and transform array <code>vector4f[key_count] transforms; // XYZW floats</code> that will correspond to the frame with the matching array index. Unlike in ACL decompressed tracks, native uncompressed tracks' <code>W</code> component is always <code>0.0</code>. Both of these element arrays will be null aligned to 16 bytes.		Each transform type will get a frame table as <code>uint16_t[key_count] keys;</code> where each value is the frame that the keyframe will be inserted to, and transform array <code>vector4f[key_count] transforms; // XYZW floats</code> that will correspond to the frame with the matching array index. Unlike in ACL decompressed tracks, native uncompressed tracks' <code>W</code> component is always <code>0.0</code>. Both of these element arrays will be null aligned to 16 bytes.

	~~===~~ Posing and Transform Inheritance: ~~===~~		Posing and Transform Inheritance:

	Each bone's transformation is stored as an object space (AKA 'Pose Space' or 'Model Space') transform relative to its parent bone's final transformation. That means, regardless of what the reset pose or bone orientations are in the model's skeleton, the stored values of any given frame will completely overwrite any pose previously present.		Each bone's transformation is stored as an object space (AKA 'Pose Space' or 'Model Space') transform relative to its parent bone's final transformation. That means, regardless of what the reset pose or bone orientations are in the model's skeleton, the stored values of any given frame will completely overwrite any pose previously present.

	However, there are two quirks that are unlike most interchange formats for 3D animations:		However, there are two quirks that are unlike most interchange formats for 3D animations:

	~~1.)~~ Scale inheritance is a simple local space copy of the parent's local space scale. There is no shearing or skewing of any kind. <~~!--~~ In Blender, a bone's scale inheritance mode can be changed from `Full` to `Aligned` to achieve this effect -->		# Scale inheritance is a simple local space copy of the parent's local space scale. There is no shearing or skewing of any kind.
			#* <sub>In Blender, a bone's scale inheritance mode can be changed from `Full` to `Aligned` to achieve this effect</sub>
	~~2.)~~ Calculating the position and rotation of any given bone is to be done independent of any scale values.		# Calculating the position and rotation of any given bone is to be done independent of any scale values.

	When reading in-game, though scales from recursive parents are inherited and multiplied, translation is completely unaffected in any capacity, unintuitively. Therefore, before applying any object space transformation in a setting where parent scale is assumed to affect the child bone's location, each bone's final scale must be calculated and applied by multiplying its and each recursive parent's scales together.		When reading in-game, though scales from recursive parents are inherited and multiplied, translation is completely unaffected in any capacity, unintuitively. Therefore, before applying any object space transformation in a setting where parent scale is assumed to affect the child bone's location, each bone's final scale must be calculated and applied by multiplying its and each recursive parent's scales together.
Line 97:		Line 98:

	''<sub>TODO...</sub>''		''<sub>TODO...</sub>''

			== Thanks ==

			* [https://github.com/Turk645 Turk645:] Original uncompressed PXD blender importer scripts
			* [https://github.com/WistfulHopes WistfulHopes:] ACL tools, blender export scripts
			* [https://github.com/ik-01 ik-01]: Format and game code research, filling in gaps for unknown values
			* [https://github.com/AdelQue AdelQue:] Format research, math relations and bone inheritance behaviors

AdelQue: Added uncompressed animation section, cleaned up

2025-04-08T07:37:46Z

Added uncompressed animation section, cleaned up

Show changes

AdelQue: Added unfinished warning

2025-04-06T01:26:55Z

Added unfinished warning

← Older revision		Revision as of 01:26, 6 April 2025
Line 1:		Line 1:
			{{Notice\|type=warn\|content=This page is unfinished.}}

	== PXD Files ==		== PXD Files ==
	Hedgehog Engine 2 games from 2016 Olympics to Shadow Generations contain skeleton and skeletal animation files with the file extensions <code>.skl.pxd</code> and <code>.anm.pxd</code> respectively, stored in a [[BINA]] container. Skeletons contain pose data that usually result in a T-Pose, bone and parent indices, and bone names. Animations contain playback metadata such as frame rate and frame count, track count, and either compressed animation pose data using [https://github.com/nfrechette/acl ACL compression,] or raw uncompressed pose data that use bone and frame indices to specify keyframed transforms, with linear interpolation between keyframes. For both skeletons and animations, position and rotation data for each bone is an absolute transform relative to the parent position, whereas scales are inherited from their respective parents locally, and have no effect on their positions.		Hedgehog Engine 2 games from 2016 Olympics to Shadow Generations contain skeleton and skeletal animation files with the file extensions <code>.skl.pxd</code> and <code>.anm.pxd</code> respectively, stored in a [[BINA]] container. Skeletons contain pose data that usually result in a T-Pose, bone and parent indices, and bone names. Animations contain playback metadata such as frame rate and frame count, track count, and either compressed animation pose data using [https://github.com/nfrechette/acl ACL compression,] or raw uncompressed pose data that use bone and frame indices to specify keyframed transforms, with linear interpolation between keyframes. For both skeletons and animations, position and rotation data for each bone is an absolute transform relative to the parent position, whereas scales are inherited from their respective parents locally, and have no effect on their positions.

AdelQue: Started PXD Documentation

2025-04-06T01:24:59Z

Started PXD Documentation

New page

== PXD Files ==
Hedgehog Engine 2 games from 2016 Olympics to Shadow Generations contain skeleton and skeletal animation files with the file extensions <code>.skl.pxd</code> and <code>.anm.pxd</code> respectively, stored in a [[BINA]] container. Skeletons contain pose data that usually result in a T-Pose, bone and parent indices, and bone names. Animations contain playback metadata such as frame rate and frame count, track count, and either compressed animation pose data using [https://github.com/nfrechette/acl ACL compression,] or raw uncompressed pose data that use bone and frame indices to specify keyframed transforms, with linear interpolation between keyframes. For both skeletons and animations, position and rotation data for each bone is an absolute transform relative to the parent position, whereas scales are inherited from their respective parents locally, and have no effect on their positions.

== <code>.anm.pxd</code> Structure ==
Animation files are stored in a [[BINA]] container. Below is the contents stored within the <code>DATA</code> node beginning at file offset <code>0x40</code>:

=== Header: ===
<syntaxhighlight lang="c++">struct Header {
char[4] magic; // 'NAXP'
uint32_t version; // Always 0x200
uint8_t additive; // 0x01 if additive, else 0x00
uint8_t compressed; // 0x08 if ACL compressed, 0x00 if uncompressed
char[6]; // 0x00 align to 8 bytes
uint64_t metadata_offset; // Offset to metadata, always 0x18
float32 duration; // Duration of the animation in seconds, calculated as ((frame_count-1) / FPS)
uint32_t frame_count; // Frame count
uint32_t track_count; // Bone count
uint64_t skel_anim_offset; // Offset for character's skeletal animation, always 0x40 if compressed, 0x38 if uncompressed
uint64_t root_anim_offset; // Offset for root motion animation, 0x00 if no root motion is present, aligned to 16 bytes
if (compressed)
char[8]; // Likely alignment to 16 bytes for ACL data. Always 0x00 if compressed
};</syntaxhighlight>

=== ACL Data: ===
All Hedgehog Engine 2 games that utilize the ACL library appear to use v2.0.0. [https://github.com/nfrechette/acl/tree/a54c5c2781be9f14b840a60f8dd8ec6c5065885d/docs See the ACL docs for more info.] If <code>compressed == 0x08</code>, skeletal animations and root motion animations are compressed.

==== Compressed Data: ====
Below is a high level overview of the stored ACL chunk data:<syntaxhighlight lang="c++">uint32_t acl_chunk_size; // Total size of chunk
int32_t acl_hash; // Hash
uint32_t acl_tag; // Identifies ACL buffer type, always 0xAC11AC11 to mark compressed_tracks
// https://github.com/nfrechette/acl/blob/976ff051048477f2281c7d3609fddf0b3cba2c2d/includes/acl/core/buffer_tag.h#L49

uint16_t acl_version; // ACL Version Enum Identifier. 0x07 indicates v2.0.0, the only version observed in any Hedgehog Engine 2 game.
// https://github.com/nfrechette/acl/blob/976ff051048477f2281c7d3609fddf0b3cba2c2d/includes/acl/core/compressed_tracks_version.h#L71

char acl_padding; // always 0x00
uint8_t acl_track_type; // ACL Track Type Enum. 0x0C indicates qvvf, the only track type observed in any Hedgehog Engine 2 game.
// https://github.com/nfrechette/acl/blob/976ff051048477f2281c7d3609fddf0b3cba2c2d/includes/acl/core/track_types.h#L68

uint32_t track_count; // Bone count if skeletal animation, 0x01 if root motion. Should match header track_count if skeletal.
uint32_t sample_count; // Frame count, should match header frame_count.
float32 sample_rate; // Playback FPS of animation, should equal header ((frame_count - 1) / duration)
char[acl_chunk_size - 0x1C] acl_data; // String of compressed ACL data</syntaxhighlight>

==== Raw/Decompressed Track List: ====
''TODO...''

=== Uncompressed Animation Data: ===
''TODO...''

== <code>.skl.pxd</code> Structure ==
Skeleton files are stored in a [[BINA]] container. Below is the contents stored within the <code>DATA</code> node beginning at file offset <code>0x40</code>:

''TODO...''