def calc_hand_contact_prob()

in scripts/data_analysis/hand_contact_prob.py [0:0]

• 30 lines of code
• 8 McCabe index (conditional complexity)

def calc_hand_contact_prob(p_nums, intents, object_names, contact_thresh=0.4,
                           search_r=15e-3, hand_idx=1):
  """
  hand_idx: 0 for left, 1 for right
  """
  contact_probs = []
  for p_num in p_nums:
    for intent in intents:
      if object_names is None:
        object_names = get_object_names(p_num, intent)
      for object_name in object_names:
        print('{:d} : {:s} : {:s}'.format(p_num, intent, object_name))
        cp = ContactPose(p_num, intent, object_name)
        object_mesh = o3dio.read_triangle_mesh(cp.contactmap_filename)
        v = np.array(object_mesh.vertices)
        c = np.array(object_mesh.vertex_colors)[:, 0]
        c = mutils.texture_proc(c)
        idx = c >= contact_thresh
        v = v[idx]
        
        # read mano
        hand = cp.mano_meshes()[hand_idx]
        if hand is None:
          continue

        h_pc = o3dg.PointCloud()
        h_pc.points = o3du.Vector3dVector(hand['vertices'])
        tree = o3dg.KDTreeFlann(h_pc)

        contact_prob = np.zeros(len(hand['vertices']))
        for vv in v:
          k, idx, dist2 = tree.search_hybrid_vector_3d(vv, search_r, 10)
          for i in range(k):
            # contact_prob[idx[i]] += (1.0/np.sqrt(dist2[i]))
            contact_prob[idx[i]] = 1
        contact_probs.append(contact_prob)
  contact_probs = np.mean(contact_probs, axis=0)
  return contact_probs