text
stringlengths
1
93.6k
feed_dict={
J_in: X_input_J,
P_in: X_input_P,
B_in: X_input_B,
})
gal_features_all_int.extend(Seq_features_int.tolist())
gal_features_all.extend(Seq_features.tolist())
gal_features_all_P.extend(Seq_features_P.tolist())
gal_features_all_B.extend(Seq_features_B.tolist())
# gal_features_tea.extend(rep_tea_.tolist())
gal_labels_all.extend(labels.tolist())
tr_step += 1
return gal_features_all_int, gal_features_all, gal_features_all_P, gal_features_all_B, gal_labels_all
def evaluation():
vl_step = 0
vl_size = X_test_J.shape[0]
pro_labels_all = []
pro_features_all = []
pro_features_tea = []
pro_features_tea_2 = []
vl_step = 0
vl_size = X_test_J.shape[0]
pro_labels_all = []
pro_features_all_int = []
pro_features_all = []
pro_features_all_P = []
pro_features_all_B = []
while vl_step * batch_size < vl_size:
if (vl_step + 1) * batch_size > vl_size:
break
X_input_J = X_test_J[vl_step * batch_size:(vl_step + 1) * batch_size]
X_input_P = X_test_P[vl_step * batch_size:(vl_step + 1) * batch_size]
X_input_B = X_test_B[vl_step * batch_size:(vl_step + 1) * batch_size]
X_input_J = X_input_J.reshape([-1, joint_num, 3])
X_input_P = X_input_P.reshape([-1, 10, 3])
X_input_B = X_input_B.reshape([-1, 5, 3])
labels = y_test[vl_step * batch_size:(vl_step + 1) * batch_size]
[Seq_features_int, Seq_features, Seq_features_P, Seq_features_B] = sess.run(
[seq_ftr_int, seq_ftr, seq_ftr_P, seq_ftr_B],
feed_dict={
J_in: X_input_J,
P_in: X_input_P,
B_in: X_input_B,
})
pro_labels_all.extend(labels.tolist())
pro_features_all_int.extend(Seq_features_int.tolist())
pro_features_all.extend(Seq_features.tolist())
pro_features_all_P.extend(Seq_features_P.tolist())
pro_features_all_B.extend(Seq_features_B.tolist())
vl_step += 1
X = np.array(gal_features_all)
X_int = np.array(gal_features_all_int)
X_P = np.array(gal_features_all_P)
X_B = np.array(gal_features_all_B)
y = np.array(gal_labels_all)
t_X = np.array(pro_features_all)
t_X_int = np.array(pro_features_all_int)
t_X_P = np.array(pro_features_all_P)
t_X_B = np.array(pro_features_all_B)
t_y = np.array(pro_labels_all)
t_y = np.argmax(t_y, axis=-1)
y = np.argmax(y, axis=-1)
def mean_ap(distmat, query_ids=None, gallery_ids=None,
query_cams=None, gallery_cams=None):
# distmat = to_numpy(distmat)
m, n = distmat.shape
# Fill up default values
if query_ids is None:
query_ids = np.arange(m)
if gallery_ids is None:
gallery_ids = np.arange(n)
if query_cams is None:
query_cams = np.zeros(m).astype(np.int32)
if gallery_cams is None:
gallery_cams = np.ones(n).astype(np.int32)
# Ensure numpy array
query_ids = np.asarray(query_ids)
gallery_ids = np.asarray(gallery_ids)
query_cams = np.asarray(query_cams)