Code for the Paper: Antonino Furnari and Giovanni Maria Farinella. What Would You Expect? Anticipating Egocentric Actions with Rolling-Unrolling LSTMs and Modality Attention. International Conference on Computer Vision, 2019.
Thank you for the pull request! I'll have a look at it and try to integrate, but it might require some time for me to be able to do so. In the meantime, feel free to update the pull request in case you want to fix something.
From a quick look (but I still have to look at it closer), it seems that the ability to pre-load features is now the default, and that the old logic to avoid pre-loading features is removed. If my understanding of your code is correct, this is not ideal. Even if 30GB is a modest amount of RAM in many modern systems, some users may still prefer the old logic. For instance, if you put the lmdb on an SSD, speed should not be an issue.
If you could edit the pull request to keep the old logic and allow the user to choose the new one by adding a flag from command line (e.g., --preload-features), it would be great and the pull request would be more easily and quickly integrated. If you cannot, don't worry, I'll try to do my best to review and edit.
Thank you for the pull request! I'll have a look at it and try to integrate, but it might require some time for me to be able to do so. In the meantime, feel free to update the pull request in case you want to fix something.
From a quick look (but I still have to look at it closer), it seems that the ability to pre-load features is now the default, and that the old logic to avoid pre-loading features is removed. If my understanding of your code is correct, this is not ideal. Even if 30GB is a modest amount of RAM in many modern systems, some users may still prefer the old logic. For instance, if you put the lmdb on an SSD, speed should not be an issue.
If you could edit the pull request to keep the old logic and allow the user to choose the new one by adding a flag from command line (e.g.,
--preload-features
), it would be great and the pull request would be more easily and quickly integrated. If you cannot, don't worry, I'll try to do my best to review and edit.Antonino