Merge pull request #10 from taokong/kongtao

resolve dist_train issues by fixing mmcv==0.2.16

docs/INSTALL.md

@@ -8,7 +8,7 @@
 - CUDA 9.0 or higher
 - NCCL 2
 - GCC 4.9 or higher
-- [mmcv](https://github.com/open-mmlab/mmcv)
+- [mmcv 0.2.16](https://github.com/open-mmlab/mmcv/tree/v0.2.16)
 
 We have tested the following versions of OS and softwares:
 

mmdet/apis/train.py

@@ -205,10 +205,7 @@ def _dist_train(model,
         for ds in dataset
     ]
     # put model on gpus
-    # model = MMDistributedDataParallel(model.cuda())
-    model = MMDistributedDataParallel(model.cuda(), 
-                                      device_ids=[torch.cuda.current_device()], 
-                                      broadcast_buffers=False)
+    model = MMDistributedDataParallel(model.cuda())
 
     # build runner
     optimizer = build_optimizer(model, cfg.optimizer)

requirements/runtime.txt

 matplotlib
-mmcv>=0.3.1
+mmcv==0.2.16
 numpy
 scipy
 # need older pillow until torchvision is fixed