RPKM/FPKM are normalization techniques used in bulk RNA-seq to mitigate library size and gene length. This is not consistant to compare between conditions as the total number of RPKM/FPKM in each condition is dissimilar. TPM normalization is slightly better as it outputs the same total number of reads in each condition. But these techniques skip over the RNA population in each samples. (Take a look at TMM or RLE normalization method fore bulk RNAseq)

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7373998/

In single cell analysis, the counts often contain high levels of technical noise with many dropouts (zero or near-zero values), the capture efficiency often varies from cell to cell, such that counts cannot be directly compared between cells. You can try to pool counts for multiple cells (by cell type for example). Take a look here

Is it right to apply some quality control checks such as count depth, number of detected genes per cell

This should be done prior RPKM/FPKM calculation, as the goal of this technique is to mitigate count depth.