diff --git a/README.md b/README.md
index 03f8163c9e9cfc792febab6f44e6911d7dc10f1d..682cbc66b5c141fc051a94447fe0b76a018f42f2 100644
--- a/README.md
+++ b/README.md
@@ -97,12 +97,12 @@ FCOS_imprv_dcnv2_X_101_64x4d_FPN_2x | Yes | 160ms | 46.6 | [download](https://cl
 
 *Note that `imprv` denotes `improvements` in our paper Table 3. These almost cost-free changes improve the performance by ~1.5% in total. Thus, we highly recommend to use them. The following are original models presented in our initial paper.*
 
-Model | Multi-scale training | Testing time / im | AP (minival) | Link
+Model | Multi-scale training | Testing time / im | AP (minival) | AP (test-dev) | Link
 --- |:---:|:---:|:---:|:---:
-FCOS_R_50_FPN_1x | No | 64ms | 37.0 | [download](https://cloudstor.aarnet.edu.au/plus/s/dDeDPBLEAt19Xrl/download)
-FCOS_R_101_FPN_2x | Yes | 78ms | 41.3 | [download](https://cloudstor.aarnet.edu.au/plus/s/vjL3L0AW7vnhRTo/download)
-FCOS_X_101_32x8d_FPN_2x | Yes | 132ms | 42.5 | [download](https://cloudstor.aarnet.edu.au/plus/s/U5myBfGF7MviZ97/download)
-FCOS_X_101_64x4d_FPN_2x | Yes | 135ms | 43.0 | [download](https://cloudstor.aarnet.edu.au/plus/s/wpwoCi4S8iajFi9/download)
+FCOS_R_50_FPN_1x | No | 64ms | 37.1 | 37.4 | [download](https://cloudstor.aarnet.edu.au/plus/s/dDeDPBLEAt19Xrl/download)
+FCOS_R_101_FPN_2x | Yes | 78ms | 41.4 | 41.5 | [download](https://cloudstor.aarnet.edu.au/plus/s/vjL3L0AW7vnhRTo/download)
+FCOS_X_101_32x8d_FPN_2x | Yes | 132ms | 42.5 | 42.7 | [download](https://cloudstor.aarnet.edu.au/plus/s/U5myBfGF7MviZ97/download)
+FCOS_X_101_64x4d_FPN_2x | Yes | 135ms | 43.0 | 43.2 | [download](https://cloudstor.aarnet.edu.au/plus/s/wpwoCi4S8iajFi9/download)
 
 **MobileNets:**