explain LR_SCHEDULE in configs (#1452)

examples/FasterRCNN/config.py

@@ -147,10 +147,12 @@ _C.TRAIN.STEPS_PER_EPOCH = 500
 _C.TRAIN.STARTING_EPOCH = 1  # the first epoch to start with, useful to continue a training
 
 # LR_SCHEDULE means equivalent steps when the total batch size is 8.
+# It can be either a string like "3x" that refers to standard convention, or a list of int.
+# LR_SCHEDULE=3x is the same as LR_SCHEDULE=[420000, 500000, 540000], which
+# means to decrease LR at steps 420k and 500k and stop training at 540k.
 # When the total bs!=8, the actual iterations to decrease learning rate, and
 # the base learning rate are computed from BASE_LR and LR_SCHEDULE.
 # Therefore, there is *no need* to modify the config if you only change the number of GPUs.
-
 _C.TRAIN.LR_SCHEDULE = "1x"      # "1x" schedule in detectron
 _C.TRAIN.EVAL_PERIOD = 50  # period (epochs) to run evaluation
 _C.TRAIN.CHECKPOINT_PERIOD = 20  # period (epochs) to save model