Update GETTING_STARTED.md

Author: jason718

docs/GETTING_STARTED.md

@@ -53,12 +53,12 @@ python -m torch.distributed.launch --nproc_per_node=$NGPUS tools/train_net.py \
 ```
 
 ### Known issues
-1. If you get error `RuntimeError: CUDA error: device-side assert triggered`, try to re-launch the program.
+1. If you get error `RuntimeError: CUDA error: device-side assert triggered`, check issue #22, or simply re-lanunch again.
 2. Since the voc datasets are very small, the best results usually are not achieved in the last epoch.
 Please save the intermediate models frequently (change `SOLVER.CHECKPOINT_PERIOD`) and check validation (`voc 2007 test`) results 
 (especially these models around the 1st learning rate dropping time). 
 You can also change the random seed `SEED` to a different value or `-1` (random).
 Note that setting `SEED` to a fixed value still cannot guarantee deterministic behavior, see explanations 
 [here](https://github.com/facebookresearch/detectron2/blob/e0e166d864a2021a15a2bc2c9234d04938066265/detectron2/config/defaults.py#L604).
 3. `OSError: Cannot allocate memory`: please check [this thread](https://github.com/prlz77/ResNeXt.pytorch/issues/5)
-4. The displayed `max mem` doesn't align with the trun memory usage. 
+4. The displayed `max mem` doesn't align with the trun memory usage.