PointTAD icon indicating copy to clipboard operation
PointTAD copied to clipboard

[NeurIPS 2022] PointTAD: Multi-Label Temporal Action Detection with Learnable Query Points

Results 5 PointTAD issues
Sort by recently updated
recently updated
newest added

請問模型產生的results_eval.csv是sparse results嗎?如果想要轉換成dense results是不是需要透過multithumos_dense.ipynb進行轉換?

您好,非常感谢您分享的代码文档,我从论文的附录中看到,pointTAD在thumos14数据集中对detection-mAP进行了评估。我已经有了thumos14数据集的.json,在进行复现的时候更改了注释文件的路径和num_classes的值,除此之外其他参数并没有修改,这种情况下复现的结果很差(如图),远达不到论文附录的性能。请问是需要调整什么其他参数或者数据吗? 望您解答! 谢谢! ![J0TZZRCIG@AQBDE)Z C7T11](https://github.com/MCG-NJU/PointTAD/assets/71828036/1a5ef847-a52c-4401-aa66-28a018d60568)

Hi, Thanks for your solid theoretical and experiment proof. But as mentioned in the paper, your work is trained in an end-to-end way while others, e.g. mlad, used pretrained i3d...

请问数据集的json是如何生成的

請問每種動作在results_eval.csv的score有general的標準嗎?