add main_prog in stnet load_vars

fluid/PaddleCV/video/models/stnet/stnet.py

@@ -132,7 +132,7 @@ class STNET(ModelBase):
                     and (not ("batch_norm" in var.name)) and (not ("xception" in var.name)) and (not ("conv3d" in var.name))
 
         vars = filter(is_parameter, prog.list_vars())
-        fluid.io.load_vars(exe, pretrain, vars=vars)
+        fluid.io.load_vars(exe, pretrain, vars=vars, main_program=prog)
 
         param_tensor = fluid.global_scope().find_var(
             "conv1_weights").get_tensor()