Mobile Net Detection Reshape Problem

[mustafa-qamaruddin]

ValueError: Dimension size must be evenly divisible by 2808 but is 588 for 'Interpre_prediction/pred_class_probs' (op: 'Reshape') with input shapes: [98,6], [3] and with input tensors computed as partial shapes: input[1] = [?,468,6].

This is the output of MobileNetDet Conv2d

      predict = slim.conv2d(dropout, num_output, kernel_size=(3, 3), stride=1, padding='SAME',
                            activation_fn=None,
                            weights_initializer=tf.truncated_normal_initializer(stddev=0.0001),
                            scope="MobileNet/conv_predict")

# probability
  batch_size = tf.shape(input_mask)[0]
  num_class_probs = config.NUM_ANCHORS * config.NUM_CLASSES
  pred_class_probs = tf.reshape(
    tf.nn.softmax(
      tf.reshape(
        prediction[:, :, :, :num_class_probs],
        [-1, config.NUM_CLASSES]
      )
    ),
    [batch_size, config.NUM_ANCHORS * fea_h * fea_w, config.NUM_CLASSES],
    name='pred_class_probs'
  )