ChrisAllenMing
commented
3 years ago Hi, thanks for your interests in our work. In cross-category video highlight detection, we consider a practical situation where we have labaled videos on source video category and unlabeled videos on target video category. In such a situation, we would like to explore how to get a good highlight detector for target category using both labeled source videos and unlabeled target videos. Therefore, for the category space ['dog', 'gymnastics', 'parkour', 'skating', 'skiing', 'surfing'], we can indeed have 65=30 different video category pairs, and we can set src_cls and tgt_cls to each of these pairs. For example, in the case of src_cls=surfing and tgt_cls=skiing*, we train the corresponding method (Source-only, DA baselines or DL-VHD) with both labeled surfing videos and unlabeled skiing videos, and finally test the model on skiing videos, which can evaluate how this method transfers the highlight knowledge from surfing to skiing.
您好 ，我在通过你的模型进行网络训练时，有一个疑惑就是关于红色框中的输入，cls src_cls tgt_cls分别代表什么呢 我看了一下源码，他们都是['dog', 'gymnastics', 'parkour', 'skating', 'skiing', 'surfing'] 列表中的其中一个，所以我没有明白Source-only Target-oracle. DA baselines.的区别是啥，或者说怎么得到的，希望您能告诉我 在运行Source-only Target-oracle. DA baselines.的Python命令时，--src_category --tgt_category 这两个参数应该分别输入什么呢？ 我觉得不可能是将 ['dog', 'gymnastics', 'parkour', 'skating', 'skiing', 'surfing']进行两两组合进行训练，这样的话会产生很多模型。