**5. Conclusions**

In this chapter, we proposed a method for the saliency estimation with weak subitizing supervision. We designed a model with the saliency subitizing process (SSP), which generates the initial saliency map using subitizing information. Without any seeds from unsupervised methods, this method outperforms other weakly supervised methods and even performs comparable to some fully supervised methods.


*probability, BCE: binary cross-entropy, GAN: generative adversarial network.*

#### **Table 5.**

*Description of saliency models.*

Finally, as this work is a first approximation, future work would be to verify how its saliency map would improve if the SUP update module were added.
