search

LeonHLJ / FAC-Net

Foreground-Action Consistency Network for Weakly Supervised Temporal Action Localization
MIT License
26 stars 5 forks source link

ActivityNet V1.3数据集的缺失问题 #9

Open mymuli opened 4 months ago

mymuli commented 4 months ago

您好！ 我在您的utils/video_dataloader.py文件里面，有看到-I3D-JOINTFeatures.npy文件：

https://github.com/LeonHLJ/FAC-Net/blob/c0108860ecb679a14ef1c3722581b20434271433/utils/video_dataloader.py#L14 我想问一下，您是否可以提供一下ActivityNet1.3-I3D-JOINTFeatures.npy文件和ActivityNet1.3-Annotation文件夹里面的文件呢？

或者我如何制作一个总的ActivityNet1.3-I3D-JOINTFeatures.np文件以及ActivityNet1.3-Annotation文件夹里面的classlist.npy、labels_all.npy等文件呢？

冒昧打扰，十分抱歉！

LeonHLJ commented 4 months ago

您好！ 我在您的utils/video_dataloader.py文件里面，有看到-I3D-JOINTFeatures.npy文件：

https://github.com/LeonHLJ/FAC-Net/blob/c0108860ecb679a14ef1c3722581b20434271433/utils/video_dataloader.py#L14

我想问一下，您是否可以提供一下ActivityNet1.3-I3D-JOINTFeatures.npy文件和ActivityNet1.3-Annotation文件夹里面的文件呢？ 或者我如何制作一个总的ActivityNet1.3-I3D-JOINTFeatures.np文件以及ActivityNet1.3-Annotation文件夹里面的classlist.npy、labels_all.npy等文件呢？

冒昧打扰，十分抱歉！

Thank you for your interest in our research. For detailed training pipeline, including the extracted features and required files, and testing strategies applicable to ActivityNet, you can refer to ACMNet.

mymuli commented 4 months ago

非常感谢您的回复！

我想问一下，您的FAC-Net模型在ActivityNet1.3数据集的实验， 是利用 ACM-Net模型作者 生成的ActivityNet1.3数据集特征和它的数据读取框架吗？

LeonHLJ commented 4 months ago

非常感谢您的回复！

我想问一下，您的FAC-Net模型在ActivityNet1.3数据集的实验， 是利用 ACM-Net模型作者 生成的ActivityNet1.3数据集特征和它的数据读取框架吗？

Yes. You can try the training and evaluation code from ACM-Net for AcNet 1.3. However, you'll need to make some modifications to the evaluation pipeline, primarily in terms of adjusting certain hyperparameters and post-processings.

mymuli commented 4 months ago

好的，非常感谢您！