SSD_training.ipynbのfit_generatorで発生したIncompatible shapesエラー

前提・実現したいこと

現在SSD_kerasの機械学習のテストとして、
https://github.com/rykov8/ssd_keras
に同梱されているSSD_training.ipynbをGoogleColabで動作させようとしています。
とりあえずepoch＝2で動作すればいいかな、と思ったのですが

発生している問題・エラーメッセージ

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-12-201d308d4729> in <module>()
      9                               validation_data=gen.generate(False),
     10                               validation_steps=gen.val_batches,
---> 11                               workers=1)

6 frames
/usr/local/lib/python3.7/dist-packages/keras/engine/training_v1.py in fit_generator(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)
   1245         use_multiprocessing=use_multiprocessing,
   1246         shuffle=shuffle,
-> 1247         initial_epoch=initial_epoch)
   1248 
   1249   def evaluate_generator(self,

/usr/local/lib/python3.7/dist-packages/keras/engine/training_v1.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, max_queue_size, workers, use_multiprocessing, **kwargs)
    794         max_queue_size=max_queue_size,
    795         workers=workers,
--> 796         use_multiprocessing=use_multiprocessing)
    797 
    798   def evaluate(self,

/usr/local/lib/python3.7/dist-packages/keras/engine/training_generator_v1.py in fit(self, model, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, max_queue_size, workers, use_multiprocessing)
    584         shuffle=shuffle,
    585         initial_epoch=initial_epoch,
--> 586         steps_name='steps_per_epoch')
    587 
    588   def evaluate(self,

/usr/local/lib/python3.7/dist-packages/keras/engine/training_generator_v1.py in model_iteration(model, data, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch, mode, batch_size, steps_name, **kwargs)
    250 
    251       is_deferred = not model._is_compiled
--> 252       batch_outs = batch_function(*batch_data)
    253       if not isinstance(batch_outs, list):
    254         batch_outs = [batch_outs]

/usr/local/lib/python3.7/dist-packages/keras/engine/training_v1.py in train_on_batch(self, x, y, sample_weight, class_weight, reset_metrics)
   1074       self._update_sample_weight_modes(sample_weights=sample_weights)
   1075       self._make_train_function()
-> 1076       outputs = self.train_function(ins)  # pylint: disable=not-callable
   1077 
   1078     if reset_metrics:

/usr/local/lib/python3.7/dist-packages/keras/backend.py in __call__(self, inputs)
   4185 
   4186     fetched = self._callable_fn(*array_vals,
-> 4187                                 run_metadata=self.run_metadata)
   4188     self._call_fetch_callbacks(fetched[-len(self._fetches):])
   4189     output_structure = tf.nest.pack_sequence_as(

/usr/local/lib/python3.7/dist-packages/tensorflow/python/client/session.py in __call__(self, *args, **kwargs)
   1483         ret = tf_session.TF_SessionRunCallable(self._session._session,
   1484                                                self._handle, args,
-> 1485                                                run_metadata_ptr)
   1486         if run_metadata:
   1487           proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

InvalidArgumentError: Incompatible shapes: [16,7308,4] vs. [16,933,4]
	 [[{{node training/Adam/gradients/gradients/loss/predictions_loss/sub_1_grad/BroadcastGradientArgs}}]]

該当のソースコード

base_lr = 3e-4
optim = keras.optimizers.Adam(lr=base_lr)

model.compile(optimizer=optim,
              loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=2.0).compute_loss)
epochs = 2
batch_size=8
history = model.fit_generator(gen.generate(True), steps_per_epoch =gen.train_batches//batch_size,
                              epochs=epochs, verbose=1,
                              callbacks=callbacks,
                              validation_data=gen.generate(False),
                              validation_steps=gen.val_batches,
                              workers=1)

model.fit_generatorは下記のqiitaのサイトを参考に変更した点ですが、元のコードも併記しておきます

history = model.fit_generator(gen.generate(True), gen.train_batches,
                              nb_epoch, verbose=1,
                              callbacks=callbacks,
                              validation_data=gen.generate(False),
                              nb_val_samples=gen.val_batches,
                              nb_worker=1)