BLEU4/CIDEr

[cyy-1102]

I'm sorry to bother you again, professor. As for the two indicators BLEU4 and CIDEr of Dense Video Caption task in the paper, how can I get these two results？

[ttengwang]

Hi, youcan try the following command.

cd PDVC/densevid_eval3
python evaluate2018.py -v -s path/to/results.json

[cyy-1102]

Hi, youcan try the following command.

cd PDVC/densevid_eval3
python evaluate2018.py -v -s path/to/results.json

I'm sorry, professor, I don't seem to understand you. I modified your code, then trained the model, and tested my trained model using 'eval.py'. In the generated 'val.log' file: METEOR, Recall, Precision, soda_c, para_Bleu_1-4, para_METEOR, para_ROUGE_L, para_CIDEr. No file to generate result.json. Can BLEU4/CIDEr be calculated directly using 'evaluate2018.py'? Don't I need to change the code while training the model?

[ttengwang]

To find the result json, you do not need to modify the code. After training, I believe you could find some JSON files in save/**/prediction/num*_epoch*_rerank*.json. These files are the generated results at each epoch.

After running eval.py, you can also find a json file in save/**/*.json which stores the generated captions of the best checkpoint.

Use evaluate2018.py to evaluate these files, you will get the score of all types of metrics.

[cyy-1102]

To find the result json, you do not need to modify the code. After training, I believe you could find some JSON files in save/**/prediction/num*_epoch*_rerank*.json. These files are the generated results at each epoch.

After running eval.py, you can also find a json file in save/**/*.json which stores the generated captions of the best checkpoint.

Use evaluate2018.py to evaluate these files, you will get the score of all types of metrics.

thank you very much, professor~