Merge pull request #1 from torrvision/master

Update from original
torrvision · Mar 31, 2016 · 54eba6b · 54eba6b
2 parents 3ddb109 + 39ded56
commit 54eba6b
Show file tree

Hide file tree

Showing 10 changed files with 75 additions and 105 deletions.
diff --git a/README.md b/README.md
@@ -3,7 +3,7 @@
 
 <b>Live demo:</b> [http://crfasrnn.torr.vision](http://crfasrnn.torr.vision)
 
-This package contains code for the "CRF-RNN" semantic image segmentation method, published in the ICCV 2015 paper [Conditional Random Fields as Recurrent Neural Networks](http://www.robots.ox.ac.uk/~szheng/papers/CRFasRNN.pdf). Our software is built on top of the [Caffe](http://caffe.berkeleyvision.org/) deep learning library. The current version was developed by:
+This package contains code for the "CRF-RNN" semantic image segmentation method, published in the ICCV 2015 paper [Conditional Random Fields as Recurrent Neural Networks](http://www.robots.ox.ac.uk/~szheng/papers/CRFasRNN.pdf). This paper was initially described in an [arXiv tech report](http://arxiv.org/abs/1502.03240). Our software is built on top of the [Caffe](http://caffe.berkeleyvision.org/) deep learning library. The current version was developed by:
 
 [Sadeep Jayasumana](http://www.robots.ox.ac.uk/~sadeep/),
 [Shuai Zheng](http://kylezheng.org/),
@@ -102,6 +102,35 @@ That's it! Enjoy our software!
 ###1.3 Run the demo
 Matlab and Python scripts for running the demo are available in the matlab-scripts and python-scripts directories, respectively. You can choose either of them. Note that you should change the paths in the scripts according your environment.
 
+####For Python fans:
+First you need to download the model. In Linux, this is:
+```
+sh download_trained_model.sh
+```
+Atlernatively, you can also get the model by directly clicking the link in python-scripts/README.md.
+
+Get into the python-scripts folder, and then type:
+```
+python crfasrnn_demo.py
+```
+You will get an output.png image.
+
+To use your own images, just replace "input.jpg" in the crfasrnn_demo.py file.
+
+####For Matlab fans:
+First you need to download the model. In Linux, this is:
+```
+sh download_trained_model.sh
+```
+Atlernatively, you can also get the model by directly clicking the link in matlab-scripts/README.md.
+
+Get into the matlab-scripts folder, load your matlab, then run the crfrnn_demo.m.
+
+To use your own images, just replace "input.jpg" in the crfrnn_demo.m file.
+
+You can also find part of our model in [MatConvNet](http://www.vlfeat.org/matconvnet/pretrained/).
+
+
 # LICENSE
 CRF-RNN feature in Caffe is implemented for the paper:
 Shuai Zheng, Sadeep Jayasumana, Bernardino Romera-Paredes, Vibhav Vineet, Zhizhong Su, Dalong Du, Chang Huang, Philip H. S. Torr.

diff --git a/caffe-crfrnn/Makefile.config b/caffe-crfrnn/Makefile.config
diff --git a/matlab-scripts/2007_000033.png b/matlab-scripts/2007_000033.png
diff --git a/matlab-scripts/README.md b/matlab-scripts/README.md
@@ -8,6 +8,11 @@ sha1: bfda5c5149d566aa56695789fa9a08e7a7f3070a
 
 This model is for the ICCV paper titled "Conditional Random Fields as Recurrent Neural Networks".
 
+With this model, you should get 73.0 mean-IOU score on a reduced validation set(346 Images http://www.robots.ox.ac.uk/~szheng/Res_CRFRNN/seg12valid.txt) on PASCAL VOC 2012. Notice that without this CRF-RNN layer, our end-to-end-trained FCN-8s model gives 69.85 mean-IOU score (mean pix. accuracy:92.94, pixel accuracy: 78.80) on the this reduced validation set of PASCAL VOC 2012.
+
+The input is expected in BGR color channel order, with the following per-channel mean substracted:
+B: 104.00698793 G: 116.66876762 R: 122.67891434
+
 Demo website is <http://crfasrnn.torr.vision>.
 
 This model was trained by 

diff --git a/matlab-scripts/crfrnn_demo.m b/matlab-scripts/crfrnn_demo.m
@@ -1,27 +1,33 @@
-clear all;
-close all;
-%This is a software bundle "CRF-RNN", which is published in a ICCV paper titled "Conditional Random Fields as Recurrent Neural Networks". This is implemented as part of the Caffe library, written in C++/C. The current version is maintained by:
+% This package contains code for the "CRF-RNN" semantic image segmentation method, published in the 
+% ICCV 2015 paper Conditional Random Fields as Recurrent Neural Networks. Our software is built on 
+% top of the Caffe deep learning library.
+% 
+% Contact:
+% Shuai Zheng (szheng@robots.ox.ac.uk), Sadeep Jayasumana (sadeep@robots.ox.ac.uk), Bernardino Romera-Paredes (bernard@robots.ox.ac.uk)
 %
-%Shuai Zheng : szheng@robots.ox.ac.uk Sadeep Jayasumana : sadeep@robots.ox.ac.uk Bernardino Romera Paredes :
+% Supervisor: 
+% Philip Torr (philip.torr@eng.ox.ac.uk)
 %
-%Supervisor: Philip Torr : philip.torr@eng.ox.ac.uk
-%
-%For more information about CRF-RNN please vist the project website http://crfasrnn.torr.vision.
+% For more information about CRF-RNN please vist the project website http://crfasrnn.torr.vision.
 %
+
 caffe_path = '../caffe-crfrnn/';
 
 model_def_file = 'TVG_CRFRNN_COCO_VOC.prototxt';
 model_file = 'TVG_CRFRNN_COCO_VOC.caffemodel';
 
-use_gpu = 1;
+if exist(model_file, 'file') ~= 2
+    error('You need a network model file. Please download our default model by running ./download_trained_model.sh');
+end
+
+use_gpu = 1; % Set this to 0 if you don't have a GPU.
 
 addpath(fullfile(caffe_path, 'matlab/caffe'));
 
 caffe('reset');
-caffe('set_device', 0);% change here if you have a powerful GPU in different device, nvidia-smi will help you check the device information.
+caffe('set_device', 0); % Change here if you have a powerful GPU in different device, nvidia-smi will help you check the device information.
 
 tvg_matcaffe_init(use_gpu, model_def_file, model_file);
-[~, map] = imread('2007_000033.png']);
 
 im = imread('input.jpg');
 
@@ -46,5 +52,5 @@
 pred = pred';
 pred = pred(1:h, 1:w);
 
-imwrite(pred, map, ['output.png'], 'png');    
-
+load map.mat
+imwrite(pred, map, 'output.png', 'png');    
diff --git a/matlab-scripts/download_trained_model.sh b/matlab-scripts/download_trained_model.sh
@@ -1 +1 @@
-wget http://goo.gl/j7PrPZ
+wget -O TVG_CRFRNN_COCO_VOC.caffemodel http://goo.gl/j7PrPZ 
diff --git a/matlab-scripts/map.mat b/matlab-scripts/map.mat
diff --git a/python-scripts/README.md b/python-scripts/README.md
@@ -6,7 +6,12 @@ license: Non-commercial, for commercial use, please contact crfasrnn@gmail.com
 sha1: bfda5c5149d566aa56695789fa9a08e7a7f3070a
 ---
 
-This model is for the ICCV paper titled "Conditional Random Fields as Recurrent Neural Networks".
+This model is for the ICCV paper titled "Conditional Random Fields as Recurrent Neural Networks". 
+
+With this model, you should get 73.0 mean-IOU score on a reduced validation set(346 Images http://www.robots.ox.ac.uk/~szheng/Res_CRFRNN/seg12valid.txt) on PASCAL VOC 2012. Notice that without this CRF-RNN layer, our end-to-end-trained FCN-8s model gives 69.85 mean-IOU score (mean pix. accuracy:92.94, pixel accuracy: 78.80) on the this reduced validation set of PASCAL VOC 2012.
+
+The input is expected in BGR color channel order, with the following per-channel mean substracted:
+B: 104.00698793 G: 116.66876762 R: 122.67891434
 
 Demo website is <http://crfasrnn.torr.vision>.
 

diff --git a/python-scripts/crfasrnn_demo.py b/python-scripts/crfasrnn_demo.py
@@ -1,17 +1,21 @@
 # -*- coding: utf-8 -*-
 """
-This is a software bundle "CRF-RNN", which is published in a ICCV paper titled "Conditional Random Fields as Recurrent Neural Networks". This is implemented as part of the Caffe library, written in C++/C. The current version is maintained by:
+This package contains code for the "CRF-RNN" semantic image segmentation method, published in the 
+ICCV 2015 paper Conditional Random Fields as Recurrent Neural Networks. Our software is built on 
+top of the Caffe deep learning library.
+ 
+Contact:
+Shuai Zheng (szheng@robots.ox.ac.uk), Sadeep Jayasumana (sadeep@robots.ox.ac.uk), Bernardino Romera-Paredes (bernard@robots.ox.ac.uk)
 
-Shuai Zheng : szheng@robots.ox.ac.uk Sadeep Jayasumana : sadeep@robots.ox.ac.uk Bernardino Romera Paredes :
+Supervisor: 
+Philip Torr (philip.torr@eng.ox.ac.uk)
 
-Supervisor: Philip Torr : philip.torr@eng.ox.ac.uk
-
-For more information about CRF-RNN please vist the project website http://crfasrnn.torr.vision.
+For more information about CRF-RNN, please vist the project website http://crfasrnn.torr.vision.
 """
 
 caffe_root = '../caffe-crfrnn/'
 import sys
-sys.path.insert(0,caffe_root+'python')
+sys.path.insert(0, caffe_root + 'python')
 
 import os
 import cPickle
@@ -27,12 +31,12 @@
 
 MODEL_FILE = 'TVG_CRFRNN_COCO_VOC.prototxt'
 PRETRAINED = 'TVG_CRFRNN_COCO_VOC.caffemodel'
-IMAGE_FILE = '2007_000032.jpg'
+IMAGE_FILE = 'input.jpg'
 
 
 #caffe.set_mode_gpu()
 net = caffe.Segmenter(MODEL_FILE, PRETRAINED)
-input_image = 255*caffe.io.load_image(IMAGE_FILE)
+input_image = 255 * caffe.io.load_image(IMAGE_FILE)
 
 
 width = input_image.shape[0]
@@ -72,7 +76,7 @@
             192,192,0]
 
 mean_vec = np.array([103.939, 116.779, 123.68], dtype=np.float32)
-reshaped_mean_vec = mean_vec.reshape(1,1,3);
+reshaped_mean_vec = mean_vec.reshape(1, 1, 3);
 
 # Rearrange channels to form BGR
 im = image[:,:,::-1]
@@ -86,10 +90,9 @@
 im = np.pad(im, pad_width=((0, pad_h), (0, pad_w), (0, 0)), mode = 'constant', constant_values = 0)
 # Get predictions
 segmentation = net.predict([im])
-segmentation2 = segmentation[0:cur_h,0:cur_w]
+segmentation2 = segmentation[0:cur_h, 0:cur_w]
 output_im = PILImage.fromarray(segmentation2)
 output_im.putpalette(pallete)
 
-
 plt.imshow(output_im)
 plt.savefig('output.png')
diff --git a/python-scripts/download_trained_model.sh b/python-scripts/download_trained_model.sh
@@ -1 +1 @@
-wget http://goo.gl/j7PrPZ TVG_CRFRNN_COCO_VOC.caffemodel
+wget -O TVG_CRFRNN_COCO_VOC.caffemodel http://goo.gl/j7PrPZ