Could we say that there is not variance mean dependence with this kind of plot  ?

From the doc :  

The red line depicts the running median estimator (window-width 10%). If there is no variance-mean dependence, then the line should be approximately horizontal.

So yes, red line is approximately horizontal but my judgement is biased....so i wanted an external point of view.

There is some dependence there but it's not bad for downstream analysis. Was this vst() or rlog()? The problem is when the median (red curve) for the far left side is in general higher than the right side. This means that the very smallest counts contribute much more to the distance metric than the high counts, and this occurs when taking log2(count + x) when x is small.