Update README.md

README.md

@@ -5,6 +5,7 @@ An unofficial PyTorch implementation of DetCo [paper](https://arxiv.org/pdf/2102
 
 This implementation is based on the official implementation of [MoCo](https://github.com/facebookresearch/moco) and [PyTorch resnet code](https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py). Though list is mentioned in original paper, this implementation use pure tensors without list, which enables parallel computing.
 
+
 ## Requirements
 ```
 pip install -r requirements.txt
@@ -14,16 +15,20 @@ pip install -r requirements.txt
 - Please refer to [main_decto.py](https://github.com/shuuchen/DetCo.pytorch/blob/main/main_decto.py):
 
 ## Unsupervised training
-- This implementation reserved original moco implementation as much as possible. Therefore the command line for unsupervised training is also similar to that of [moco](https://github.com/facebookresearch/moco#unsupervised-training). You can train like:
-```python
-python main_detco.py \
-  -a resnet50 \
-  --lr 0.015 \
-  --batch-size 128 \
-  --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \
-  [your imagenet-folder with train and val folders] \
-  --detco-t 0.2 --aug-plus --cos
-```
+As specified in the original paper, DetCo is trained with 8 GPUs with batch shuffling. However, sometimes you cannot use all the GPUs in an 8-GPU machine. I used half of them for training. If you also need to modify the number of GPUs like me, check the following two lines of code:
+  - [number of GPUs to use](https://github.com/shuuchen/DetCo.pytorch/blob/main/main_detco.py#L124)
+  - [master GPU number](https://github.com/shuuchen/DetCo.pytorch/blob/main/main_detco.py#L136)
+
+This implementation reserved original moco implementation as much as possible. Therefore the command line for unsupervised training is also similar to that of [moco](https://github.com/facebookresearch/moco#unsupervised-training). You can train like:
+  - ```python
+    python main_detco.py \
+      -a resnet50 \
+      --lr 0.015 \
+      --batch-size 128 \
+      --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \
+      [your imagenet-folder with train and val folders] \
+      --detco-t 0.2 --aug-plus --cos
+    ```
 
 ## TODO
 - [ ] object detection training & performance checking on certain datasets