Getting one word as caption with zero probability using pretrained checkpoints for image captioning-im2txt

Question

I am using pre-trained 2M checkpoints for im2txt. I am getting one word as captions as below. I am using tensorflow version 1.0 Ubuntu 16.04

0)     (p=0.000000)
1)the  (p=0.000000)
2)two  (p=0.000000)

I am expecting a complete sentence but got an irrelevant one-word caption

Answer 1

Under caption_generator.py

change the following line as follow

        most_likely_words = np.argsort(word_probabilities)[-self.beam_size:][::-1]

If the problem persists, try using a checkpoints file that has been updated with the newest code.