[FasterRCNN] update resnet101 results

examples/FasterRCNN/README.md

@@ -41,7 +41,7 @@ Predict on an image (and show output in a window):
 ```
 
 Evaluate the performance of a model and save to json.
-(A pretrained model can be downloaded in [model zoo](http://models.tensorpack.com/FasterRCNN):
+(Pretrained models can be downloaded in [model zoo](http://models.tensorpack.com/FasterRCNN):
 ```
 ./train.py --evaluate output.json --load /path/to/model
 ```
@@ -57,8 +57,10 @@ MaskRCNN results contain both bbox and segm mAP.
 | Res50 | 256 | (600, 1024) | 34.4 | 49h on 8 TitanX |
 | Res50 | 512 | (800, 1333) | 35.6 | 55h on 8 P100|
 | Res50 | 512 | (800, 1333) | 36.9/32.3 | 59h on 8 P100|
+| Res101 | 512 | (800, 1333) | 40.1/34.4 | 70h on 8 P100|
 
-Note that these models are trained with a longer learning schedule than the paper.
+Note that these models are trained with a longer learning schedule than the paper,
+and get about 1mAP better performance.
 
 ## Notes
 

examples/FasterRCNN/config.py

@@ -5,22 +5,23 @@
 import numpy as np
 
 # mode flags ---------------------
-MODE_MASK = False
+MODE_MASK = True
 
 # dataset -----------------------
 BASEDIR = '/path/to/your/COCO/DIR'
 TRAIN_DATASET = ['train2014', 'valminusminival2014']
-VAL_DATASET = 'minival2014'   # only support evaluation on one dataset
+VAL_DATASET = 'minival2014'   # only support evaluation on single dataset
 NUM_CLASS = 81
 CLASS_NAMES = []  # NUM_CLASS strings
 
 # basemodel ----------------------
 RESNET_NUM_BLOCK = [3, 4, 6, 3]     # resnet50
+# RESNET_NUM_BLOCK = [3, 4, 23, 3]     # resnet101
 
 # preprocessing --------------------
-SHORT_EDGE_SIZE = 600
-MAX_SIZE = 1024
-# alternative (better) setting: 800, 1333
+SHORT_EDGE_SIZE = 800
+MAX_SIZE = 1333
+# alternative (worse & faster) setting: 600, 1024
 
 # anchors -------------------------
 ANCHOR_STRIDE = 16