Which neural network architecture to use to detect very close and very small blobs in high resolution fluorescence images?

Question

Context

I am developing a pipeline to automate the detection of small, almost circular, bright blobs (4px) (see first image below) on high-resolution fluorescence images (2048px) and later to assign them to cells. The goal is to compute the number of blobs per cell

Data set

The training examples (in their simplest form) are 2-channel x 2048 x 2048 px images together with the position of the foci. The first channel contains the cell nuclei (see last image below) and the second channel is a marker for the blobs (see second image below).

Questions

1. I would like to know what kind of neural network architecture addresses best the problem
2. Which loss function to use
3. Which metric to report

Requirements

1. The speed is not an issue
2. It must discern blobs 1-2 pixels apart.

Thanks in advance for your opinions :)