question

[wangben123456]

I am very surprised that you did a really good thing.But,when I use your pps code in my computer ,the results is really bad,I want to know why it is.
| Rank1,mIoU 0.3351 | Rank1,IoU@0.1 0.7311 | Rank1,IoU@0.3 0.4804 | Rank1,IoU@0.5 0.2735 | Rank5,mIoU 0.5172 | Rank5,IoU@0.1 0.9160 | Rank5,IoU@0.3 0.7259 | Rank5,IoU@0.5 0.5399 |
epoch30
I am hoping for your answer.Thank you very much.

[sunoh-kim]

Thank you for your interest and for sharing your results.
The performance you mentioned at epoch 30 is not the best result, and this is the case for many other video grounding algorithms as well, since accuracy can vary significantly across epochs. It’s important to check the full trend rather than relying on a single checkpoint.

You can monitor the exact performance trends using Weights & Biases (wandb), where all training metrics are logged for detailed comparison.

For more detailed results and discussion, please refer to the following closed issue: #3