Merge pull request #2 from AmitMY/patch-1

simplify pose2text processor

multimodalhugs/processors/pose2text_preprocessor.py

@@ -17,7 +17,7 @@
 )
 from multimodalhugs.processors import MultimodalSecuence2TextTranslationProcessor
 from pose_format import Pose
-from pose_format.utils.generic import reduce_holistic, pose_hide_legs
+from pose_format.utils.generic import reduce_holistic, pose_hide_legs, pose_normalization_info
 
 logger = logging.getLogger(__name__)
 
@@ -37,21 +37,18 @@ def __init__(
         super().__init__(tokenizer=tokenizer, **kwargs)
 
     def _pose_file_to_tensor(self, pose_file: Union[str, Path]):
-        pose_file = open(pose_file, "rb").read()
-        pose = Pose.read(pose_file) # [t, people, d, xyz]
-
-        P1 = ("POSE_LANDMARKS", "RIGHT_SHOULDER") if pose.header.components[0].name == "POSE_LANDMARKS" else ("BODY_135", "RShoulder")
-        P2 = ("POSE_LANDMARKS", "LEFT_SHOULDER") if pose.header.components[0].name == "POSE_LANDMARKS" else ("BODY_135", "LShoulder")
-
+        with open(pose_file, "rb") as pose_file:
+            pose = Pose.read(pose_file.read()) # [t, people, d, xyz]
+
         pose_hide_legs(pose)
 
         if self.reduce_holistic_poses:
+            # This will be skipped if the pose is not holistic
             pose = reduce_holistic(pose) # [t, people, d', xyz]
 
-        pose = pose.normalize(pose.header.normalization_info(p1=P1, p2=P2))
-        pose = pose.torch().body.data.squeeze(1) # [t, d', xyz]
-        pose = pose.reshape(shape=(pose.shape[0], -1))
-        return pose.zero_filled()
+        pose = pose.normalize(pose_normalization_info(pose.header))
+        tensor = pose.torch().body.data.zero_filled()
+        return tensor.contiguous().view(tensor.size(0), -1)
 
     def _obtain_multimodal_input_and_masks(self, batch, **kwargs):
         tensor_secuences = [self._pose_file_to_tensor(sample["source"]) for sample in batch]