Merge pull request #311 from oarriaga/human_pose_3D

Human pose 3 d
oarriaga · Aug 28, 2023 · 12ce71f · 12ce71f
2 parents 0be29a2 + 1ccb0ca
commit 12ce71f
Show file tree

Hide file tree

Showing 4 changed files with 26 additions and 24 deletions.
diff --git a/paz/backend/image/draw.py b/paz/backend/image/draw.py
@@ -465,15 +465,15 @@ def draw_human_pose6D(image, rotation, translation, camaera_intrinsics):
                          [0, 0, 1]])
     points2D = project_to_image(rotation, translation,
                                 points3D, camaera_intrinsics)
-    points2D = points2D.astype(np.int)
+    points2D = points2D.astype(np.int32)
 
     x = points2D[0]
     y = points2D[1]
     z = points2D[2]
 
-    x_hat = (x / np.linalg.norm(x) * 60).astype(np.int)
-    y_hat = (y / np.linalg.norm(y) * 60).astype(np.int)
-    z_hat = (z / np.linalg.norm(z) * 60).astype(np.int)
+    x_hat = (x / np.linalg.norm(x) * 60).astype(np.int32)
+    y_hat = (y / np.linalg.norm(y) * 60).astype(np.int32)
+    z_hat = (z / np.linalg.norm(z) * 60).astype(np.int32)
 
     offset = [50, 50]
     image = draw_line(image, offset, x_hat + offset,

diff --git a/paz/backend/keypoints.py b/paz/backend/keypoints.py
@@ -640,7 +640,7 @@ def merge_into_mean(keypoints2D, args_to_mean):
     # Returns:
              keypoints2D: keypoints2D after merging
             """
-    keypoints = keypoints2D.copy()
+    keypoints = np.array(keypoints2D.copy())
     for point, joints_indices in args_to_mean.items():
         keypoints[:, point] = (keypoints[:, joints_indices[0]] +
                                keypoints[:, joints_indices[1]]) / 2

diff --git a/paz/pipelines/keypoints.py b/paz/pipelines/keypoints.py
@@ -434,6 +434,8 @@ def __init__(self, solver, camera_intrinsics,
                  args_to_joints3D=args_to_joints3D, filter=True, draw=True,
                  draw_pose=True):
         super(EstimateHumanPose, self).__init__()
+        self.pose3D = []
+        self.pose6D = []
         self.draw = draw
         self.filter = filter
         self.draw_pose = draw_pose
@@ -448,17 +450,18 @@ def __init__(self, solver, camera_intrinsics,
 
     def call(self, image):
         inferences2D = self.estimate_keypoints_2D(image)
-        keypoints2D = np.array(inferences2D['keypoints'])
+        keypoints2D = inferences2D['keypoints']
         if self.draw:
             image = inferences2D['image']
-        keypoints3D = self.estimate_keypoints_3D(keypoints2D)
-        keypoints3D = np.reshape(keypoints3D, (-1, 32, 3))
-        optimized_output = self.optimize(keypoints3D, keypoints2D)
-        joints2D, joints3D, pose3D, projection2D = optimized_output
-        pose6D = human_pose3D_to_pose6D(pose3D[0])
-        if self.draw_pose:
-            rotation, translation = pose6D
-            image = self.draw_pose6D(image, rotation, translation)
-            translation = ["%.2f" % item for item in translation]
-            image = self.draw_text(image, str(translation), (30, 30))
-        return self.wrap(image, keypoints2D, pose3D, pose6D)
+        if len(keypoints2D) > 0:
+            keypoints3D = self.estimate_keypoints_3D(keypoints2D)
+            keypoints3D = np.reshape(keypoints3D, (-1, 32, 3))
+            optimized_output = self.optimize(keypoints3D, keypoints2D)
+            joints2D, joints3D, self.pose3D, projection2D = optimized_output
+            self.pose6D = human_pose3D_to_pose6D(self.pose3D[0])
+            if self.draw_pose:
+                rotation, translation = self.pose6D
+                image = self.draw_pose6D(image, rotation, translation)
+                translation = ["%.2f" % item for item in translation]
+                image = self.draw_text(image, str(translation), (30, 30))
+        return self.wrap(image, keypoints2D, self.pose3D, self.pose6D)
diff --git a/paz/processors/keypoints.py b/paz/processors/keypoints.py
@@ -25,7 +25,6 @@
 from ..datasets.human36m import human_start_joints
 
 
-
 class ProjectKeypoints(Processor):
     """Projects homogenous keypoints (4D) in the camera coordinates system into
         image coordinates using a projective transformation.
@@ -337,16 +336,16 @@ def __init__(self, args_to_joints3D, solver, camera_intrinsics):
 
     def call(self, keypoints3D, keypoints2D):
         joints3D = filter_keypoints3D(keypoints3D, self.args_to_joints3D)
-        filtered_keypoints2D = self.filter_keypoints2D(keypoints2D)
-        root2D = filtered_keypoints2D[:, :2]
+        joints2D = self.filter_keypoints2D(keypoints2D)
+        root2D = joints2D[:, :2]
         length2D, length3D = get_bones_length(
-            filtered_keypoints2D, keypoints3D, human_start_joints)
+            joints2D, keypoints3D, human_start_joints)
         ratio = length3D / length2D
         initial_joint_translation = initialize_translation(
             root2D, self.camera_intrinsics, ratio)
         joint_translation = solve_least_squares(
             self.solver, compute_reprojection_error, initial_joint_translation,
-            joints3D, filtered_keypoints2D, self.camera_intrinsics)
-        optimized_poses3D, projected_points2D = compute_optimized_pose3D(
+            joints3D, joints2D, self.camera_intrinsics)
+        optimized_poses3D, projection2D = compute_optimized_pose3D(
             keypoints3D, joint_translation, self.camera_intrinsics)
-        return filtered_keypoints2D, joints3D, optimized_poses3D, projected_points2D
+        return joints2D, joints3D, optimized_poses3D, projection2D