I'm running a Deseq2 with RNA seq data to compare between test and control which every condition has 3 duplicated. But when I compare between test vs. control I got 1797 DEG but control against test I got 1760 genes.

res<-results(dds, contrast = c("dex", "T0", "T20"),alpha = 0.05)
res2 <- lfcShrink(dds=dds, coef="dex_T0_vs_T20")
sig <- res2[ which(res2$padj < 0.05), ]
summary(sig)

out of 1797 with nonzero total read count
adjusted p-value < 0.05
LFC > 0 (up)       : 952, 53%
LFC < 0 (down)     : 845, 47%

res<-results(dds, contrast = c("dex", "T20", "T0"),alpha = 0.05)
res2 <- lfcShrink(dds=dds, coef="dex_T20_vs_T0")
sig <- res2[ which(res2$padj < 0.05), ]
summary(sig)

out of 1760 with nonzero total read count
adjusted p-value < 0.1
LFC > 0 (up)       : 929, 53%
LFC < 0 (down)     : 831, 47%


sessionInfo( )
R version 4.0.2 (2020-06-22)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19042)

Matrix products: default

locale:
[1] LC_COLLATE=Thai_Thailand.874  LC_CTYPE=Thai_Thailand.874   
[3] LC_MONETARY=Thai_Thailand.874 LC_NUMERIC=C                 
[5] LC_TIME=Thai_Thailand.874    

attached base packages:
[1] parallel  stats4    stats     graphics  grDevices utils     datasets 
[8] methods   base     

other attached packages:
 [1] RColorBrewer_1.1-2          Biostrings_2.56.0          
 [3] XVector_0.28.0              stringr_1.4.0              
 [5] dplyr_1.0.2                 DESeq2_1.28.1              
 [7] SummarizedExperiment_1.18.2 DelayedArray_0.14.1        
 [9] matrixStats_0.57.0          Biobase_2.48.0             
[11] GenomicRanges_1.40.0        GenomeInfoDb_1.24.2        
[13] IRanges_2.22.2              S4Vectors_0.26.1           
[15] BiocGenerics_0.34.0        

loaded via a namespace (and not attached):
 [1] bdsmatrix_1.3-4        Rcpp_1.0.5             locfit_1.5-9.4        
 [4] mvtnorm_1.1-1          apeglm_1.10.0          lattice_0.20-41       
 [7] digest_0.6.25          plyr_1.8.6             R6_2.4.1              
[10] emdbook_1.3.12         coda_0.19-4            RSQLite_2.2.1         
[13] ggplot2_3.3.2          pillar_1.4.6           zlibbioc_1.34.0       
[16] rlang_0.4.7            rstudioapi_0.11        annotate_1.66.0       
[19] blob_1.2.1             Matrix_1.2-18          bbmle_1.0.23.1        
[22] splines_4.0.2          BiocParallel_1.22.0    geneplotter_1.66.0    
[25] RCurl_1.98-1.2         bit_4.0.4              munsell_0.5.0         
[28] tinytex_0.26           numDeriv_2016.8-1.1    compiler_4.0.2        
[31] xfun_0.18              pkgconfig_2.0.3        tidyselect_1.1.0      
[34] tibble_3.0.3           GenomeInfoDbData_1.2.3 XML_3.99-0.5          
[37] crayon_1.3.4           MASS_7.3-51.6          bitops_1.0-6          
[40] grid_4.0.2             xtable_1.8-4           gtable_0.3.0          
[43] lifecycle_0.2.0        DBI_1.1.0              magrittr_1.5          
[46] scales_1.1.1           stringi_1.5.3          genefilter_1.70.0     
[49] ellipsis_0.3.1         vctrs_0.3.4            generics_0.0.2        
[52] tools_4.0.2            bit64_4.0.5            glue_1.4.2            
[55] purrr_0.3.4            survival_3.1-12        AnnotationDbi_1.50.3  
[58] colorspace_1.4-1       memoise_1.1.0

There can be slight variation when you relevel in a design with more than a two level factor. I recommend to keep the reference level set for comparisons involving the reference level, and only relevel to compare against non-reference level conditions.