search

facebookresearch / HierVL

[CVPR 2023] HierVL Learning Hierarchical Video-Language Embeddings
Other
43 stars 3 forks source link

Loss Equations in Paper #12

Closed yeyingdege closed 2 months ago

yeyingdege commented 3 months ago

Hi, thank you very much for the excellent work for solving long video representation learning problem. However, there might be a small typo in the loss equations in the manuscript, which is a missing "-" in the equations.

thechargedneutron commented 3 months ago

Hi @yeyingdege , thanks for reaching out. We mean to maximize the given training objective. So, the overall objective is to max L(child) and max L(parent). In other words, as you said it should be min - L(child) and min - L(parent). Sorry, if this was confusing and thanks for seeking a clarification. The training code takes care of this.

thechargedneutron commented 2 months ago

Closing this issue, feel free to reopen if you have further questions.