search

ttengwang / PDVC

End-to-End Dense Video Captioning with Parallel Decoding (ICCV 2021)
MIT License
200 stars 23 forks source link

i3d+vggish results #27

Closed upccpu closed 2 years ago

upccpu commented 2 years ago

Hello professor,

When I reproduced your ‘i3d+vggish’ model, I found that i can not achieve the same results compared with the original paper. I don't know if there is something wrong with my settings.

Thinks

ttengwang commented 2 years ago

Could you please show me some detailed results?

upccpu commented 2 years ago

Thanks for rely.

i list my final results below, it was captured from the 17 epoch. By the way, i didn't change the configrations in anet_c3d_pdvc.yml and anet_i3dvgg_pdvc.yml in the training procedure.

Validation results of iter 165006: Bleu_1:0.14397184343673516 Bleu_2:0.07563979885710821 Bleu_3:0.03841529336432101 Bleu_4:0.018592074153146686 METEOR:0.07360392698576934 ROUGE_L:0.14248773409982254 CIDEr:0.2695501870140006 Recall:0.49363335351742516 Precision:0.5138465188800716 soda_c:0.05815482000571441 para_Bleu_1:0.4600347481728046 para_Bleu_2:0.2734444234216087 para_Bleu_3:0.16531830682605833 para_Bleu_4:0.10265513451045642 para_METEOR:0.14629477337476632 para_ROUGE_L:0.28967922834328586 para_CIDEr:0.17420500091591917

overall score of iter 165006: 0.13175874699148377

ttengwang commented 2 years ago

The results are normal since all scores you mentioned above are based on incomplete validation videos (~4483 out of 4917). Around 434 videos got a METEOR of 0. The METEOR on 4483 videos is 7.36039 * 4917/4483 = ~8.07

upccpu commented 2 years ago

Thanks for your guidance. You wrote about this issue in the paper, but I ignored it. I realized this problem after you reminded me.