minor change (Nathan's comment)

collections/_posts/2023-11-22-DETR.md

@@ -99,7 +99,7 @@ Once the best permutation $$\hat{\sigma}_i$$ is found, the *Hungarian loss* is c
 
 $$ \mathcal{L}_{Hungarian}(y, \hat{y}) = \sum_{i=1}^{N} [ - \log \hat{p}_{\hat{\sigma} (i)} (c_i) + \unicode{x1D7D9}_{ \{ c_i \notin \emptyset \} } \mathcal{L}_{box} (b_i, \hat{b}_{\hat{\sigma} (i)}) ]$$
 
-The authors found helpful to use auxiliary loss inside the decoder during training. The added prediction FFNs (N.B: all the FFNs share the same parameters) and Hungarian loss *after each decoder layer*. 
+The authors found helpful to use additional losses inside the decoder during training. They added prediction FFNs (N.B: all the FFNs share the same parameters) and Hungarian loss *after each decoder layer*. 
 
 # Results