Replaced the "DAC" resistors with an association to get more precision. The association values have been calculated using an online tool (link)

Nevertheless, the shades yet were not gradual at the center. After some tweaking adding resistors in parallel I've managed to figure out which resistor in the network was disturbing the "linearity" of the output. It happened that to be the 195 Ohms resistor association. Few experiments later I have added a 120 Ohm resistor in parallel thus resulting in 75 ohms (sic) but I've managed to get a good image.

It didn't occurred to me at the moment but maybe there is one bit flipped either in software or in the circuit.