diff --git a/README.md b/README.md
index 3d88779..65043c8 100644
--- a/README.md
+++ b/README.md
@@ -1,14 +1,23 @@
 # Mask RCNN
 Mask RCNN in TensorFlow
-This repo attempts to reproduce this amazing work by Kaiming He.
-[Mask RCNN](https://arxiv.org/abs/1703.06870).
+
+This repo attempts to reproduce this amazing work by Kaiming He et al. : 
+[Mask R-CNN](https://arxiv.org/abs/1703.06870)
+
+## Requirements
+
+- [Tensorflow (>= 1.0.0)](https://www.tensorflow.org/install/install_linux)
+- [Numpy](https://github.com/numpy/numpy/blob/master/INSTALL.rst.txt)
+- [COCO dataset](http://mscoco.org/dataset/#download)
+- [Resnet50](http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz)
 
 ## How-to
-1. Download [COCO](http://mscoco.org/dataset/#download) dataset, place it into `./data`, then run `python download_and_convert_data.py` to build tf-records. It takes a while.
-2. Download pretrained resnet50 model, `wget http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz`, unzip it, place it into `./data/pretrained_models/`
-3. Go to `./libs` and run `make`
-4. run `python train/train.py` for training
-5. There are certainly some bugs, please report them back, and let's solve them together.
+1. Go to `./libs/datasets/pycocotools` and run `make`
+2. Download [COCO](http://mscoco.org/dataset/#download) dataset, place it into `./data`, then run `python download_and_convert_data.py` to build tf-records. It takes a while.
+3. Download pretrained resnet50 model, `wget http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz`, unzip it, place it into `./data/pretrained_models/`
+4. Go to `./libs` and run `make`
+5. run `python train/train.py` for training
+6. There are certainly some bugs, please report them back, and let's solve them together.
 
 ## TODO:
 - [x] ROIAlign
@@ -29,4 +38,12 @@ This repo attempts to reproduce this amazing work by Kaiming He.
 - Anything helps this repo, including **discussion**, **testing**, **promotion** and of course **your awesome code**.
 
 ## Acknowledgment
-This repo borrows tons of code from [TFFRCNN](https://github.com/CharlesShang/TFFRCNN), [py-faster-rcnn](https://github.com/rbgirshick/py-faster-rcnn), [faster_rcnn](https://github.com/ShaoqingRen/faster_rcnn), [tf-models](https://github.com/tensorflow/models)
+This repo borrows tons of code from 
+- [TFFRCNN](https://github.com/CharlesShang/TFFRCNN)
+- [py-faster-rcnn](https://github.com/rbgirshick/py-faster-rcnn) 
+- [faster_rcnn](https://github.com/ShaoqingRen/faster_rcnn)
+- [tf-models](https://github.com/tensorflow/models)
+
+## License
+See [LICENSE](https://github.com/CharlesShang/FastMaskRCNN/blob/master/LICENSE) for details.
+
diff --git a/data/README.md b/data/README.md
index e0871de..6f09a56 100644
--- a/data/README.md
+++ b/data/README.md
@@ -1,12 +1,9 @@
-Place your coco in this dir, like
+Place and unzip your coco in this dir, like
 
 ```buildoutcfg
 ./data
     ./coco
-        ./train2014.zip
-        ./val2014.zip
-        ./instances_train-val2014.zip
-        ./person_keypoints_trainval2014.zip
-        ./captions_train-val2014.zip
-
+        ./annotations
+        ./train2014
+        ./val2014
 ```
diff --git a/libs/boxes/bbox_transform.py b/libs/boxes/bbox_transform.py
index fadd90c..e8c74e9 100644
--- a/libs/boxes/bbox_transform.py
+++ b/libs/boxes/bbox_transform.py
@@ -58,8 +58,12 @@ def bbox_transform_inv(boxes, deltas):
 
     pred_ctr_x = dx * widths[:, np.newaxis] + ctr_x[:, np.newaxis]
     pred_ctr_y = dy * heights[:, np.newaxis] + ctr_y[:, np.newaxis]
-    pred_w = np.exp(dw) * widths[:, np.newaxis]
-    pred_h = np.exp(dh) * heights[:, np.newaxis]
+    # pred_w = np.exp(dw) * widths[:, np.newaxis]
+    # pred_h = np.exp(dh) * heights[:, np.newaxis]
+
+    pred_w = np.exp(dw + np.log(widths[:, np.newaxis]))
+    pred_h = np.exp(dh + np.log(heights[:, np.newaxis]))
+
 
     pred_boxes = np.zeros(deltas.shape, dtype=deltas.dtype)
     # x1
diff --git a/train/train.py b/train/train.py
index 839c178..cb04d8f 100644
--- a/train/train.py
+++ b/train/train.py
@@ -127,8 +127,8 @@ def train():
 
     ## network
     logits, end_points, pyramid_map = network.get_network(FLAGS.network, image,
-            weight_decay=FLAGS.weight_decay)
-    outputs = pyramid_network.build(end_points, ih, iw, pyramid_map, 
+            weight_decay=FLAGS.weight_decay, is_training=True)
+    outputs = pyramid_network.build(end_points, im_shape[1], im_shape[2], pyramid_map,
             num_classes=81,
             base_anchors=9,
             is_training=True,