# Understanding the reconstruction loss in the paper “Anomaly Detection using Deep Learning based Image Completion”

I would like to implement the approach represented in this paper. Here they used following reconstruction loss:

$$L(X)= \frac{\lambda \cdot || M \odot (X - F(\overline{M} \odot X)) ||_{1} + (1 - \lambda) \cdot || \overline{M} \odot (X - F(\overline{M} \odot X)) ||_{1}}{N}$$

Unfortunately, the author does not explain the function $$F$$. Does someone know a similar function or could understand the function's purpose from the context?

$$F$$ in this context is the output of the Convolutional Neural Network that's being trained, which is of the same size as $$X$$.
• That makes total sense, because $\hat{y}$ is missing in the equation. Thank you. – oezguensi Jul 18 at 15:22