Master

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.
+

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
 ```

download_and_convert_data.py

@@ -4,6 +4,7 @@
 from __future__ import division
 from __future__ import print_function
 
+import os
 import tensorflow as tf
 
 from libs.datasets import download_and_convert_coco
@@ -21,6 +22,8 @@
 
 
 def main(_):
+  if not os.path.isdir('./output/mask_rcnn'):
+    os.makedirs('./output/mask_rcnn')
   if not FLAGS.dataset_name:
     raise ValueError('You must supply the dataset name with --dataset_name')
   if not FLAGS.dataset_dir:

libs/boxes/bbox_transform.py

@@ -58,8 +58,16 @@ 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_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

libs/configs/config_v1.py

@@ -235,7 +235,7 @@
     'Only regions which intersection is larger than fg_threshold are considered to be fg')
 
 tf.app.flags.DEFINE_float(
-    'fg_threshold', 0.5,
+    'fg_threshold', 0.7,
     'Only regions which intersection is larger than fg_threshold are considered to be fg')
 
 tf.app.flags.DEFINE_float(

libs/datasets/dataset_factory.py

@@ -2,8 +2,11 @@
 from __future__ import division
 from __future__ import print_function
 
+import tensorflow as tf
+from libs.visualization.summary_utils import visualize_input
 import glob
 from libs.datasets import coco
+
 import libs.preprocessings.coco_v1 as coco_preprocess
 
 def get_dataset(dataset_name, split_name, dataset_dir, 
@@ -16,6 +19,7 @@ def get_dataset(dataset_name, split_name, dataset_dir,
     image, ih, iw, gt_boxes, gt_masks, num_instances, img_id = coco.read(tfrecords)
 
     image, gt_boxes, gt_masks = coco_preprocess.preprocess_image(image, gt_boxes, gt_masks, is_training)
+    #visualize_input(gt_boxes, image, tf.expand_dims(gt_masks, axis=3))
 
     return image, ih, iw, gt_boxes, gt_masks, num_instances, img_id
 

libs/datasets/download_and_convert_coco.py

@@ -34,6 +34,7 @@
 tf.app.flags.DEFINE_boolean('vis',  False,
                           'Show some visual masks')
 
+
 def download_and_uncompress_zip(zip_url, dataset_dir):
   """Downloads the `zip_url` and uncompresses it locally.
      From: https://github.com/tensorflow/models/blob/master/slim/datasets/dataset_utils.py