update docs

docs/tutorial/performance-tuning.md

 
 # Performance Tuning
 
-Here's a list of things you can do when your training is slow:
+Here's a list of things you can do when your training is slow.
+And if you're going to open an issue about slow training, PLEASE do them and include your findings.
 
 ## Figure out the bottleneck
 

examples/FasterRCNN/README.md

@@ -44,10 +44,13 @@ To evaluate the performance (pretrained models can be downloaded in [model zoo](
 
 ## Results
 
-Mean Average Precision @IoU=0.50:0.95:
+Trained on trainval35k and evaluated on minival, got the following results:
+mAP@IoU=0.50:0.95:
 
-+ trainval35k/minival, FASTRCNN_BATCH=256: 34.2. Takes 49h on 8 TitanX.
-+ trainval35k/minival, FASTRCNN_BATCH=64: 33.0. Takes 22h on 8 P100.
+|Backbone | `FASTRCNN_BATCH` | mAP | Time |
+| - | - | - | - |
+| Res50 | 256 | 34.4 | 49h on 8 TitanX |
+| Res50 | 64 | 33.0 | 22h on 8 P100 |
 
 The hyperparameters are not carefully tuned. You can probably get better performance by e.g. training longer.
 

examples/FasterRCNN/config.py

@@ -17,6 +17,7 @@ RESNET_NUM_BLOCK = [3, 4, 6, 3]     # resnet50
 # preprocessing --------------------
 SHORT_EDGE_SIZE = 600
 MAX_SIZE = 1024
+# alternative (better) setting: 800, 1333
 
 # anchors -------------------------
 ANCHOR_STRIDE = 16