train_val.py 文件源码

def train_net(network, imdb, roidb, valroidb, output_dir, tb_dir,
              pretrained_model=None,
              max_iters=40000):
  """Train a Faster R-CNN network."""
  roidb = filter_roidb(roidb)
  valroidb = filter_roidb(valroidb)

  tfconfig = tf.ConfigProto(allow_soft_placement=True)
  tfconfig.gpu_options.allow_growth = True

  with tf.Session(config=tfconfig) as sess:
    sw = SolverWrapper(sess, network, imdb, roidb, valroidb, output_dir, tb_dir,
                       pretrained_model=pretrained_model)
    print('Solving...')
    sw.train_model(sess, max_iters)
    print('done solving')