1
0
mirror of https://github.com/DarklightGames/io_scene_psk_psa.git synced 2025-02-13 08:32:33 +01:00
io_scene_psk_psa/psa/reader.py

118 lines
4.8 KiB
Python
Raw Normal View History

import ctypes
import numpy as np
from .data import *
def _try_fix_cue4parse_issue_103(sequences) -> bool:
# Detect if the file was exported from CUE4Parse prior to the fix for issue #103.
# https://github.com/FabianFG/CUE4Parse/issues/103
# The issue was that the frame_start_index was not being set correctly, and was always being set to the same value
# as the frame_count.
# This fix will eventually be deprecated as it is only necessary for files exported prior to the fix.
2024-02-02 12:28:45 -08:00
if len(sequences) > 0 and sequences[0].frame_start_index == sequences[0].frame_count:
# Manually set the frame_start_index for each sequence. This assumes that the sequences are in order with
# no shared frames between sequences (all exporters that I know of do this, so it's a safe assumption).
frame_start_index = 0
for i, sequence in enumerate(sequences):
sequence.frame_start_index = frame_start_index
frame_start_index += sequence.frame_count
return True
return False
class PsaReader(object):
2024-03-14 19:04:12 -07:00
"""
This class reads the sequences and bone information immediately upon instantiation and holds onto a file handle.
2022-08-06 23:52:18 -07:00
The keyframe data is not read into memory upon instantiation due to its potentially very large size.
To read the key data for a particular sequence, call :read_sequence_keys.
2024-03-14 19:04:12 -07:00
"""
def __init__(self, path):
self.keys_data_offset: int = 0
self.fp = open(path, 'rb')
self.psa: Psa = self._read(self.fp)
@property
def bones(self):
return self.psa.bones
@property
def sequences(self):
return self.psa.sequences
def read_sequence_data_matrix(self, sequence_name: str) -> np.ndarray:
2024-03-14 19:04:12 -07:00
"""
Reads and returns the data matrix for the given sequence.
@param sequence_name: The name of the sequence.
@return: An FxBx7 matrix where F is the number of frames, B is the number of bones.
2024-03-14 19:04:12 -07:00
"""
sequence = self.psa.sequences[sequence_name]
keys = self.read_sequence_keys(sequence_name)
bone_count = len(self.bones)
matrix_size = sequence.frame_count, bone_count, 7
matrix = np.zeros(matrix_size)
keys_iter = iter(keys)
for frame_index in range(sequence.frame_count):
for bone_index in range(bone_count):
matrix[frame_index, bone_index, :] = list(next(keys_iter).data)
return matrix
def read_sequence_keys(self, sequence_name: str) -> List[Psa.Key]:
2024-03-14 19:04:12 -07:00
"""
2022-08-06 23:52:18 -07:00
Reads and returns the key data for a sequence.
@param sequence_name: The name of the sequence.
@return: A list of Psa.Keys.
2024-03-14 19:04:12 -07:00
"""
# Set the file reader to the beginning of the keys data
sequence = self.psa.sequences[sequence_name]
data_size = sizeof(Psa.Key)
bone_count = len(self.psa.bones)
buffer_length = data_size * bone_count * sequence.frame_count
sequence_keys_offset = self.keys_data_offset + (sequence.frame_start_index * bone_count * data_size)
self.fp.seek(sequence_keys_offset, 0)
buffer = self.fp.read(buffer_length)
offset = 0
keys = []
for _ in range(sequence.frame_count * bone_count):
key = Psa.Key.from_buffer_copy(buffer, offset)
keys.append(key)
offset += data_size
return keys
@staticmethod
2022-08-06 23:52:18 -07:00
def _read_types(fp, data_class, section: Section, data):
buffer_length = section.data_size * section.data_count
buffer = fp.read(buffer_length)
offset = 0
for _ in range(section.data_count):
data.append(data_class.from_buffer_copy(buffer, offset))
offset += section.data_size
def _read(self, fp) -> Psa:
psa = Psa()
while fp.read(1):
fp.seek(-1, 1)
section = Section.from_buffer_copy(fp.read(ctypes.sizeof(Section)))
if section.name == b'ANIMHEAD':
pass
elif section.name == b'BONENAMES':
PsaReader._read_types(fp, Psa.Bone, section, psa.bones)
elif section.name == b'ANIMINFO':
sequences = []
PsaReader._read_types(fp, Psa.Sequence, section, sequences)
# Try to fix CUE4Parse bug, if necessary.
_try_fix_cue4parse_issue_103(sequences)
for sequence in sequences:
psa.sequences[sequence.name.decode()] = sequence
elif section.name == b'ANIMKEYS':
# Skip keys on this pass. We will keep this file open and read from it as needed.
self.keys_data_offset = fp.tell()
fp.seek(section.data_size * section.data_count, 1)
else:
fp.seek(section.data_size * section.data_count, 1)
print(f'Unrecognized section in PSA: "{section.name}"')
return psa