diff --git a/projects/DML/README.md b/projects/DML/README.md
index 17aef17..9b2206f 100644
--- a/projects/DML/README.md
+++ b/projects/DML/README.md
@@ -1 +1,14 @@
-# Deep mutual learning
\ No newline at end of file
+# Deep mutual learning
+
+This repo implements [Deep Mutual Learning](https://arxiv.org/abs/1706.00384) (DML) for person re-id.
+
+We used this code in our [OSNet](https://arxiv.org/pdf/1905.00953.pdf) paper (see Supp. B). The training command to reproduce the result of "triplet + DML" (Table 12f in the paper) is
+```bash
+python main.py \
+--config-file im_osnet_x1_0_dml_256x128_amsgrad_cosine.yaml \
+--root $DATA
+```
+
+Change `model.deploy` to `both` if you wanna enable model ensembling.
+
+If you have any questions, please raise an issue in the Issues area.
\ No newline at end of file
diff --git a/projects/DML/im_osnet_x1_0_dml_256x128_amsgrad_cosine.yaml b/projects/DML/im_osnet_x1_0_dml_256x128_amsgrad_cosine.yaml
index e11ae5f..3a240b1 100644
--- a/projects/DML/im_osnet_x1_0_dml_256x128_amsgrad_cosine.yaml
+++ b/projects/DML/im_osnet_x1_0_dml_256x128_amsgrad_cosine.yaml
@@ -1,6 +1,7 @@
 model:
   name: 'osnet_x1_0'
   pretrained: True
+  deploy: 'model1'
 
 data:
   type: 'image'