keras-yolo3: TypeError: Cannot interpret feed_dict key as Tensor

two step to solve this problem: (1) modify the “init(self, **kwargs)” method: def init(self, **kwargs): self.dict.update(self._defaults) # set up default values self.dict.update(kwargs) # and update with user overrides self.class_names = self._get_class() self.anchors = self._get_anchors() self.sess = K.get_session() with self.sess.as_default(): with self.sess.graph.as_default(): self.K_learning_phase = K.learning_phase() self.boxes, self.scores, self.classes = self.generate()

(2) modify the “detect_image(self, image)” method: out_boxes, out_scores, out_classes = self.sess.run( [self.boxes, self.scores, self.classes], feed_dict={ self.yolo_model.input: image_data, self.input_image_shape: [image.size[1], image.size[0]], self.K_learning_phase: 0 })

done.

you can comment out this code “keras_learning_phase:0” , it’s ok to run, I have tried

In my case, I encountered the same error when creating the session (i.e., creating a YOLO class variable) from a main thread and running the session (i.e., calling YOLO.detect_image()) from a different thread in a callback function. In this case when using multiple threads, you have to set the default graph in the non-main threads before running detection. The change below fixed the issue for me. (I am running tensorflow-gpu==1.14 with keras==2.2.4)

Note: some references state that you also have to set the session again, but in my case I found that setting the default graph was sufficient to prevent the error. I have included but commented out the line to set the session in case anyone else finds it necessary.

with self.sess.graph.as_default():     # new line (required)
	#K.tensorflow_backend.set_session(self.sess)    # new line (wasn't required for me)

	out_boxes, out_scores, out_classes = self.sess.run(
		[self.boxes, self.scores, self.classes],
		feed_dict={
			self.yolo_model.input: image_data,
			self.input_image_shape: [image.size[1], image.size[0]],
			K.learning_phase(): 0
		})