update readme (add speed with torchscript)

README.md

@@ -22,14 +22,13 @@ We also propose an attention variant RCDA to reduce the memory cost for high-res
 | SMCA | multi-level | 50 | 43.7 | 152 | 10 |
 | Deformable DETR | multi-level | 50 | 43.8 | 173 | 15 |
 | Conditional DETR | DC5 | 50 | 43.8 | 195 | 10 |
-| Anchor DETR | DC5 | 50 | 44.2 | 151 | 16   |
+| Anchor DETR | DC5 | 50 | 44.2 | 151 | 16 (19) |
 
 
 *Note:*
 1. The results are based on ResNet-50 backbone.
 2. Inference speeds are measured on NVIDIA Tesla V100 GPU.
-3. DETR gets 12 FPS with torchscript optimization and 10 FPS without it.
-The others do not enable the torchscript.
+3. The values in parentheses of the Infer Speed indicate the speed with torchscript optimization.
 
 
 ## Model