search

aipixel / AEMatter

Another matter.
GNU General Public License v2.0
52 stars 2 forks source link

composition loss #6

Closed astro-fits closed 8 months ago

astro-fits commented 8 months ago

The loss function mentioned in your paper has three components, and it appears that all of these losses are calculated based on the ground truth (GT) alpha map and the predicted alpha map. My question is, why is the composition loss ignored in your work?

Thank you for your attention.

Windaway commented 8 months ago

Thank you for your question and for your attention to our work. You're correct in noting that the composition loss is a commonly used loss function in image matting tasks. However, in our specific study, we chose not to include it. We found that the values of the composition loss tended to be relatively small compared to the other components. This led us to prioritize other losses that had a more significant impact on the overall performance of the model. Again, we appreciate your interest in our work and your thoughtful question. If you have any further inquiries or if there's anything else you'd like to discuss, please feel free to let us know. Thank you.

astro-fits commented 8 months ago

Thank you so much for your quick reply.