Open CallMeRabbit opened 9 months ago
I also have the problem
I also have the problem
Yes, I also encountered the same problem. The model did not perform well on the validation machine. I tried using other convolutional networks to extract image features and pre-trained word embedding vectors, but the results were similar and did not improve. log0.txt
The model was trained on flicker8k, but the results achieved only half the BLEU-4 score mentioned by the authors (about 0.14-0.15). I have not modified any parameters in train.py. May I ask why such a result occurs?
![屏幕截图 2023-11-24 101110](https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Image-Captioning/assets/96768788/66a24e56-52f9-4bc2-ae03-0cd64569c79b)