How can I reproduce the results on MSRVTT reported in Paper?

[11362p]

Hi,Thanks for sharing your work.I tried to reproduce the results on the MSRVTT dataset and I found that the results were lower than the paper's results.The results of training the model under the official MSRVTT division method are as follows: Epoch[19 / 50] LR: 5.163553898972419e-06

• Text to Video:
• r_1_5_10, medr, meanr: [27.2, 55.9, 67.1, 4.0, 27.1]
• recall sum: 150.2
• mAP: 0.4061

• ---

• Video to text:
• r_1_5_10, medr, meanr: [44.5, 75.1, 83.9, 2.0, 10.0]
• recall sum: 203.4
• mAP: 0.2622

• ---

• Current perf: 353.6619718309859
• Best perf: 358.7625754527163

Early stopping happended.

best performance on validation: 358.7625754527163 Then, Load model_best.pth, the results obtained on the test set are as follows: write into: /VisualSearch/msrvtt10k/results/cv_tpami_2021/msrvtt10ktrain/msrvtt10k/runs_8/model_best.pth.tar

• Text to Video:
• r_1_5_10, medr, meanr: [11.4, 30.3, 41.0, 18.0, 145.9]
• recall sum: 82.7
• mAP: 0.2093

• ---

• Video to text:
• r_1_5_10, medr, meanr: [21.4, 45.7, 58.4, 7.0, 42.8]
• recall sum: 125.5
• mAP: 0.1001

• ---

  the results reported in the paper are as follows:

Could you give some tips about it? Many thanks