diff --git a/.idea/SynthText.iml b/.idea/SynthText.iml
deleted file mode 100644
index 6711606..0000000
--- a/.idea/SynthText.iml
+++ /dev/null
@@ -1,11 +0,0 @@
-
-
-
-
-
-
-
-
-
-
-
\ No newline at end of file
diff --git a/.idea/misc.xml b/.idea/misc.xml
deleted file mode 100644
index c139e1c..0000000
--- a/.idea/misc.xml
+++ /dev/null
@@ -1,4 +0,0 @@
-
-
-
-
\ No newline at end of file
diff --git a/.idea/modules.xml b/.idea/modules.xml
deleted file mode 100644
index 1757dd2..0000000
--- a/.idea/modules.xml
+++ /dev/null
@@ -1,8 +0,0 @@
-
-
-
-
-
-
-
-
\ No newline at end of file
diff --git a/.idea/vcs.xml b/.idea/vcs.xml
deleted file mode 100644
index 94a25f7..0000000
--- a/.idea/vcs.xml
+++ /dev/null
@@ -1,6 +0,0 @@
-
-
-
-
-
-
\ No newline at end of file
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..f7ca6b1
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,20 @@
+FROM python:3.7-buster
+
+# For japanese
+RUN apt-get update && apt-get install -y ffmpeg libsm6 libxext6 git libmecab2 libmecab-dev mecab mecab-ipadic mecab-ipadic-utf8 mecab-utils vim \
+ && apt-get clean \
+ && rm -rf /var/lib/apt/lists/*
+
+# jupyter notebook libs
+RUN pip install jupyterlab
+RUN pip install traitlets==5.1.1
+RUN pip install "ipykernel<5.5.2"
+
+WORKDIR /workspace
+COPY ./ /workspace/
+RUN python -m pip install -r /workspace/requirements.txt
+
+ENV TZ=Asia/Singapore
+RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
+
+WORKDIR /workspace
\ No newline at end of file
diff --git a/README.md b/README.md
index 3b9ed06..6c0d616 100644
--- a/README.md
+++ b/README.md
@@ -28,7 +28,9 @@ Add support for chinese
-The library is written in Python. The main dependencies are:
+The code in the `master` branch is for Python2. Python3 is supported in the `python3` branch.
+
+The main dependencies are:
```
pygame, opencv (version 3.3), PIL (Image), numpy, matplotlib, h5py, scipy
@@ -119,20 +121,29 @@ For an explanation of the fields in `dset.h5` (e.g.: `seg`,`area`,`label`), plea
### Pre-processed Background Images
The 8,000 background images used in the paper, along with their segmentation and depth masks, have been uploaded here:
-`http://zeus.robots.ox.ac.uk/textspot/static/db/`, where, `` can be:
+`http://www.robots.ox.ac.uk/~vgg/data/scenetext/preproc/`, where, `` can be:
+
+| filenames | size | description | md5 hash |
+|:--------------- | ----:|:---------------------------------------------------- |:-------------------------------- |
+| `imnames.cp` | 180K | names of images which do not contain background text | |
+| `bg_img.tar.gz` | 8.9G | images (filter these using `imnames.cp`) | 3eac26af5f731792c9d95838a23b5047 |
+| `depth.h5` | 15G | depth maps | af97f6e6c9651af4efb7b1ff12a5dc1b |
+| `seg.h5` | 6.9G | segmentation maps | 1605f6e629b2524a3902a5ea729e86b2 |
+
+Note: due to large size, `depth.h5` is also available for download as 3-part split-files of 5G each.
+These part files are named: `depth.h5-00, depth.h5-01, depth.h5-02`. Download using the path above, and put them together using `cat depth.h5-0* > depth.h5`.
-- `imnames.cp` [180K]: names of filtered files, i.e., those files which do not contain text
-- `bg_img.tar.gz` [8.9G]: compressed image files (more than 8000, so only use the filtered ones in imnames.cp)
-- `depth.h5` [15G]: depth maps
-- `seg.h5` [6.9G]: segmentation maps
+[`use_preproc_bg.py`](https://github.com/ankush-me/SynthText/blob/master/use_preproc_bg.py) provides sample code for reading this data.
Note: I do not own the copyright to these images.
### Generating Samples with Text in non-Latin (English) Scripts
-@JarveeLee has modified the pipeline for generating samples with Chinese text [here](https://github.com/JarveeLee/SynthText_Chinese_version).
-@gachiemchiep has modified the pipeline for generating samples with Japanese text [here](https://github.com/gachiemchiep/SynthText).
-
+- @JarveeLee has modified the pipeline for generating samples with Chinese text [here](https://github.com/JarveeLee/SynthText_Chinese_version).
+- @adavoudi has modified it for arabic/persian script, which flows from right-to-left [here](https://github.com/adavoudi/SynthText).
+- @MichalBusta has adapted it for a number of languages (e.g. Bangla, Arabic, Chinese, Japanese, Korean) [here](https://github.com/MichalBusta/E2E-MLT).
+- @gachiemchiep has adapted for Japanese [here](https://github.com/gachiemchiep/SynthText).
+- @gungui98 has adapted for Vietnamese [here](https://github.com/gungui98/SynthText).
+- @youngkyung has adapted for Korean [here](https://github.com/youngkyung/SynthText_kr).
### Further Information
Please refer to the paper for more information, or contact me (email address in the paper).
-
diff --git a/colorize3_poisson.py b/colorize3_poisson.py
index befd4e7..3cad231 100644
--- a/colorize3_poisson.py
+++ b/colorize3_poisson.py
@@ -6,11 +6,12 @@
import scipy.ndimage.interpolation as sii
import os
import os.path as osp
-import cPickle as cp
+#import cPickle as cp
+import _pickle as cp
#import Image
from PIL import Image
from poisson_reconstruct import blit_images
-
+import pickle
def sample_weighted(p_dict):
ps = p_dict.keys()
@@ -38,14 +39,18 @@ def __init__(self,alpha,color):
elif color.ndim==3: #rgb image
self.color = color.copy().astype('uint8')
else:
- print color.shape
+ print (color.shape)
raise Exception("color datatype not understood")
class FontColor(object):
def __init__(self, col_file):
- with open(col_file,'r') as f:
- self.colorsRGB = cp.load(f)
+ with open(col_file,'rb') as f:
+ #self.colorsRGB = cp.load(f)
+ u = pickle._Unpickler(f)
+ u.encoding = 'latin1'
+ p = u.load()
+ self.colorsRGB = p
self.ncol = self.colorsRGB.shape[0]
# convert color-means from RGB to LAB for better nearest neighbour
@@ -402,7 +407,7 @@ def check_perceptible(self, txt_mask, bg, txt_bg):
diff = np.linalg.norm(bg_px-txt_px,ord=None,axis=1)
diff = np.percentile(diff,[10,30,50,70,90])
- print "color diff percentile :", diff
+ print ("color diff percentile :", diff)
return diff, (bgo,txto)
def color(self, bg_arr, text_arr, hs, place_order=None, pad=20):
@@ -425,7 +430,7 @@ def color(self, bg_arr, text_arr, hs, place_order=None, pad=20):
# initialize the placement order:
if place_order is None:
- place_order = np.array(xrange(len(text_arr)))
+ place_order = np.array(range(len(text_arr)))
rendered = []
for i in place_order[::-1]:
diff --git a/common.py b/common.py
index b217734..4982573 100644
--- a/common.py
+++ b/common.py
@@ -25,17 +25,17 @@ def colorprint(colorcode, text, o=sys.stdout, bold=False):
o.write(colorize(colorcode, text, bold=bold))
def warn(msg):
- print colorize(Color.YELLOW, msg)
+ print (colorize(Color.YELLOW, msg))
def error(msg):
- print colorize(Color.RED, msg)
+ print (colorize(Color.RED, msg))
# http://stackoverflow.com/questions/366682/how-to-limit-execution-time-of-a-function-call-in-python
class TimeoutException(Exception): pass
@contextmanager
def time_limit(seconds):
def signal_handler(signum, frame):
- raise TimeoutException, colorize(Color.RED, " *** Timed out!", highlight=True)
+ raise TimeoutException(colorize(Color.RED, " *** Timed out!", highlight=True))
signal.signal(signal.SIGALRM, signal_handler)
signal.alarm(seconds)
try:
diff --git a/gen.py b/gen.py
index 2f4cd2f..76ef6eb 100644
--- a/gen.py
+++ b/gen.py
@@ -1,5 +1,5 @@
#!/usr/bin/env python
-#-*- coding: utf-8 -*-
+# -*- coding: utf-8 -*-
# Author: Ankush Gupta
# Date: 2015
@@ -37,15 +37,15 @@
OUT_FILE = 'results/SynthText.h5'
OUT_DIR = 'results'
+
def get_data():
"""
- Download the image,depth and segmentation data:
- Returns, the h5 database.
- """
+ Download the image,depth and segmentation data:
+ Returns, the h5 database.
+ """
if not osp.exists(DB_FNAME):
try:
- colorprint(Color.BLUE, '\tdownloading data (56 M) from: ' + DATA_URL, bold=True)
- print
+ print('\tdownloading data (56 M) from: ' + DATA_URL)
sys.stdout.flush()
out_fname = 'data.tar.gz'
wget.download(DATA_URL, out=out_fname)
@@ -53,10 +53,10 @@ def get_data():
tar.extractall()
tar.close()
os.remove(out_fname)
- colorprint(Color.BLUE, '\n\tdata saved at:' + DB_FNAME, bold=True)
+ print('\n\tdata saved at:' + DB_FNAME)
sys.stdout.flush()
except:
- print colorize(Color.RED, 'Data not found and have problems downloading.', bold=True)
+ print('Data not found and have problems downloading.')
sys.stdout.flush()
sys.exit(-1)
# open the h5 file and return:
@@ -65,19 +65,18 @@ def get_data():
def add_res_to_db(imgname, res, db):
"""
- Add the synthetically generated text image instance
- and other metadata to the dataset.
- """
+ Add the synthetically generated text image instance
+ and other metadata to the dataset.
+ """
ninstance = len(res)
- for i in xrange(ninstance):
+ for i in range(ninstance):
dname = "%s_%d" % (imgname, i)
db['data'].create_dataset(dname, data=res[i]['img'])
db['data'][dname].attrs['charBB'] = res[i]['charBB']
db['data'][dname].attrs['wordBB'] = res[i]['wordBB']
-
+ db['data'][dname].attrs['txt'] = res[i]['txt']
text_utf8 = [char.encode('utf8') for char in res[i]['txt']]
- db['data'][dname].attrs['txt'] = text_utf8
-
+ db['data'][dname].attrs['txt_utf8'] = text_utf8
def save_res_to_imgs(imgname, res):
"""
@@ -85,23 +84,22 @@ def save_res_to_imgs(imgname, res):
and other metadata to the dataset.
"""
ninstance = len(res)
- for i in xrange(ninstance):
+ for i in range(ninstance):
filename = "{}/{}_{}.png".format(OUT_DIR, imgname, i)
# Swap bgr to rgb so we can save into image file
img = res[i]['img'][..., [2, 1, 0]]
cv2.imwrite(filename, img)
-
def main(viz=False):
# open databases:
- print colorize(Color.BLUE, 'getting data..', bold=True)
+ print('getting data..')
db = get_data()
- print colorize(Color.BLUE, '\t-> done', bold=True)
+ print('\t-> done')
# open the output h5 file:
- out_db = h5py.File(OUT_FILE,'w')
+ out_db = h5py.File(OUT_FILE, 'w')
out_db.create_group('/data')
- print colorize(Color.GREEN,'Storing the output in: '+OUT_FILE, bold=True)
+ print('Storing the output in: ' + OUT_FILE)
# get the names of the image files in the dataset:
imnames = sorted(db['image'].keys())
@@ -112,7 +110,7 @@ def main(viz=False):
start_idx, end_idx = 0, min(NUM_IMG, N)
RV3 = RendererV3(DATA_PATH, max_time=SECS_PER_IMG, lang=args.lang)
- for i in xrange(start_idx, end_idx):
+ for i in range(start_idx, end_idx):
imname = imnames[i]
try:
# get the image:
@@ -130,23 +128,23 @@ def main(viz=False):
# re-size uniformly:
sz = depth.shape[:2][::-1]
- img = np.array(img.resize(sz, Image.ANTIALIAS))
- seg = np.array(Image.fromarray(seg).resize(sz, Image.NEAREST))
+ img = np.array(img.resize(sz, Image.Resampling.LANCZOS))
+ seg = np.array(Image.fromarray(seg).resize(sz, Image.Resampling.NEAREST))
- print colorize(Color.RED, '%d of %d' % (i, end_idx - 1), bold=True)
+ print('%d of %d' % (i, end_idx - 1))
res = RV3.render_text(img, depth, seg, area, label,
ninstance=INSTANCE_PER_IMAGE, viz=viz)
if len(res) > 0:
# non-empty : successful in placing text:
- add_res_to_db(imname,res,out_db)
+ add_res_to_db(imname, res, out_db)
# visualize the output:
if viz:
save_res_to_imgs(imname, res)
- if 'q' in raw_input(colorize(Color.RED, 'continue? (enter to continue, q to exit): ', True)):
+ if 'q' in input('continue? (enter to continue, q to exit): '):
break
except:
traceback.print_exc()
- print colorize(Color.GREEN, '>>>> CONTINUING....', bold=True)
+ print('>>>> CONTINUING....')
continue
db.close()
out_db.close()
@@ -155,10 +153,10 @@ def main(viz=False):
if __name__ == '__main__':
import argparse
- parser = argparse.ArgumentParser(description='Genereate Synthetic Scene-Text Images')
+ parser = argparse.ArgumentParser(description='Generate Synthetic Scene-Text Images')
parser.add_argument('--viz', action='store_true', dest='viz', default=False,
help='flag for turning on visualizations')
parser.add_argument('--lang', default='ENG',
- help='Select language : ENG/JPN')
+ help='Select language : ENG or JPN')
args = parser.parse_args()
main(args.viz)
diff --git a/invert_font_size.py b/invert_font_size.py
index ee70eaa..fb08179 100644
--- a/invert_font_size.py
+++ b/invert_font_size.py
@@ -18,9 +18,8 @@
models = {} # linear model
FS = FontState()
-# plt.figure()
-# plt.hold(True)
-for i in xrange(len(FS.fonts)):
+
+for i in range(len(FS.fonts)):
font = freetype.Font(FS.fonts[i], size=12)
h = []
for y in ys:
diff --git a/poisson_reconstruct.py b/poisson_reconstruct.py
index b49eab8..59f1453 100644
--- a/poisson_reconstruct.py
+++ b/poisson_reconstruct.py
@@ -99,7 +99,7 @@ def blit_images(im_top,im_back,scale_grad=1.0,mode='max'):
im_res = np.zeros_like(im_top)
# frac of gradients which come from source:
- for ch in xrange(im_top.shape[2]):
+ for ch in range(im_top.shape[2]):
ims = im_top[:,:,ch]
imd = im_back[:,:,ch]
@@ -203,7 +203,7 @@ def contiguous_regions(mask):
# plt.imshow(im_alpha_L)
# plt.show()
- for i in xrange(500,im_alpha_L.shape[1],5):
+ for i in range(500,im_alpha_L.shape[1],5):
l_actual = im_actual_L[i,:]#-im_actual_L[i,:-1]
l_alpha = im_alpha_L[i,:]#-im_alpha_L[i,:-1]
l_poisson = im_poisson_L[i,:]#-im_poisson_L[i,:-1]
@@ -212,7 +212,6 @@ def contiguous_regions(mask):
with sns.axes_style("darkgrid"):
plt.subplot(2,1,2)
plt.plot(l_alpha,label='alpha')
- plt.hold(True)
plt.plot(l_poisson,label='poisson')
plt.plot(l_actual,label='actual')
plt.legend()
@@ -227,7 +226,6 @@ def contiguous_regions(mask):
with sns.axes_style("white"):
plt.subplot(2,1,1)
plt.imshow(im_alpha[:,:,::-1].astype('uint8'))
- plt.hold(True)
plt.plot([0,im_alpha_L.shape[0]-1],[i,i],'r')
plt.axis('image')
plt.show()
diff --git a/prep_scripts/floodFill.py b/prep_scripts/floodFill.py
index 1f61ef1..10596f6 100644
--- a/prep_scripts/floodFill.py
+++ b/prep_scripts/floodFill.py
@@ -13,11 +13,11 @@
import h5py
import os.path as osp
import multiprocessing as mp
-import traceback, sys
+
def get_seed(sx,sy,ucm):
n = sx.size
- for i in xrange(n):
+ for i in range(n):
if ucm[sx[i]+1,sy[i]+1] == 0:
return (sy[i],sx[i])
@@ -41,7 +41,7 @@ def get_mask(ucm,viz=False):
sx,sy = np.where(mask==0)
seed = get_seed(sx,sy,ucm)
i += 1
- print " > terminated in %d steps"%i
+ print (" > terminated in %d steps"%i)
if viz:
plt.imshow(mask)
@@ -81,7 +81,7 @@ def get_imname(self,i):
return "".join(map(chr, self.ucm_h5[self.ucm_h5['names'][0,self.i]][:]))
def __stop__(self):
- print "DONE"
+ print ("DONE")
self.ucm_h5.close()
raise StopIteration
@@ -101,14 +101,14 @@ def get_valid_name(self):
def next(self):
imname = self.get_valid_name()
- print "%d of %d"%(self.i+1,self.N)
+ print ("%d of %d"%(self.i+1,self.N))
ucm = self.ucm_h5[self.ucm_h5['ucms'][0,self.i]][:]
ucm = ucm.copy()
self.i += 1
return ((ucm>self.th).astype('uint8'),imname)
ucm_iter = ucm_iterable(db_path,th)
- print "cpu count: ", mp.cpu_count()
+ print ("cpu count: ", mp.cpu_count())
parpool = mp.Pool(4)
ucm_result = parpool.imap_unordered(get_mask_parallel, ucm_iter, chunksize=1)
@@ -116,16 +116,16 @@ def next(self):
if res is None:
continue
((mask,area,label),imname) = res
- print "got back : ", imname
+ print ("got back : ", imname)
mask = mask.astype('uint16')
mask_dset = dbo_mask.create_dataset(imname, data=mask)
mask_dset.attrs['area'] = area
mask_dset.attrs['label'] = label
# close the h5 files:
- print "closing DB"
+ print ("closing DB")
dbo.close()
- print ">>>> DONE"
+ print (">>>> DONE")
base_dir = '/home/' # directory containing the ucm.mat, i.e., output of run_ucm.m
diff --git a/ransac.py b/ransac.py
index cb5b1fc..5bd271d 100644
--- a/ransac.py
+++ b/ransac.py
@@ -30,7 +30,7 @@ def fit_plane_ransac(pts, neighbors=None,z_pos=None, dist_inlier=0.05,
"""
n,_ = pts.shape
ninlier,models = [],[]
- for i in xrange(max_iter):
+ for i in range(max_iter):
if neighbors is None:
p = pts[np.random.choice(pts.shape[0],nsample,replace=False),:]
else:
@@ -43,7 +43,7 @@ def fit_plane_ransac(pts, neighbors=None,z_pos=None, dist_inlier=0.05,
models.append(m)
if models == []:
- print "RANSAC plane fitting failed!"
+ print ("RANSAC plane fitting failed!")
return #None
else: #refit the model to inliers:
ninlier = np.array(ninlier)
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..f105c63
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,13 @@
+cycler
+h5py
+matplotlib
+opencv-python
+nltk
+numpy
+Pillow
+pygame
+pyparsing
+pytz
+scipy
+six
+wget
diff --git a/synth_utils.py b/synth_utils.py
index ed8adc4..b517c61 100644
--- a/synth_utils.py
+++ b/synth_utils.py
@@ -233,6 +233,7 @@ def ssc(v):
"""
Returns the skew-symmetric cross-product matrix corresponding to v.
"""
+ v += np.finfo("float").eps
v /= np.linalg.norm(v)
return np.array([[ 0, -v[2], v[1]],
[ v[2], 0, -v[0]],
@@ -271,7 +272,7 @@ def unrotate2d(pts):
elif R[1,1]<0:
R[:,1] *= -1
else:
- print "Rotation matrix not understood"
+ print ("Rotation matrix not understood")
return
if R[0,0]<0 and R[1,1]<0:
R *= -1
diff --git a/synthgen.py b/synthgen.py
index 75e5aa2..fe303df 100644
--- a/synthgen.py
+++ b/synthgen.py
@@ -91,7 +91,7 @@ def filter(seg,area,label):
# filter bad regions:
filt = np.array(filt)
area = area[filt]
- R = [R[i] for i in xrange(len(R)) if filt[i]]
+ R = [R[i] for i in range(len(R)) if filt[i]]
# sort the regions based on areas:
aidx = np.argsort(-area)
@@ -111,7 +111,7 @@ def sample_grid_neighbours(mask,nsample,step=3):
y_m,x_m = np.where(mask)
mask_idx = np.zeros_like(mask,'int32')
- for i in xrange(len(y_m)):
+ for i in range(len(y_m)):
mask_idx[y_m[i],x_m[i]] = i
xp,xn = np.zeros_like(mask), np.zeros_like(mask)
@@ -136,7 +136,7 @@ def sample_grid_neighbours(mask,nsample,step=3):
Y = np.transpose(np.c_[ys,ys+s,ys-s,ys+s,ys-s][:,:,None],(1,2,0))
sample_idx = np.concatenate([Y,X],axis=1)
mask_nn_idx = np.zeros((5,sample_idx.shape[-1]),'int32')
- for i in xrange(sample_idx.shape[-1]):
+ for i in range(sample_idx.shape[-1]):
mask_nn_idx[:,i] = mask_idx[sample_idx[:,:,i][:,0],sample_idx[:,:,i][:,1]]
return mask_nn_idx
@@ -215,9 +215,9 @@ def get_text_placement_mask(xyz,mask,plane,pad=2,viz=False):
REGION : DICT output of TextRegions.get_regions
PAD : number of pixels to pad the placement-mask by
"""
- _, contour,hier = cv2.findContours(mask.copy().astype('uint8'),
+ contour,hier = cv2.findContours(mask.copy().astype('uint8'),
mode=cv2.RETR_CCOMP,
- method=cv2.CHAIN_APPROX_SIMPLE)
+ method=cv2.CHAIN_APPROX_SIMPLE)[-2:]
contour = [np.squeeze(c).astype('float') for c in contour]
#plane = np.array([plane[1],plane[0],plane[2],plane[3]])
H,W = mask.shape[:2]
@@ -226,7 +226,7 @@ def get_text_placement_mask(xyz,mask,plane,pad=2,viz=False):
pts,pts_fp = [],[]
center = np.array([W,H])/2
n_front = np.array([0.0,0.0,-1.0])
- for i in xrange(len(contour)):
+ for i in range(len(contour)):
cnt_ij = contour[i]
xyz = su.DepthCamera.plane2xyz(center, cnt_ij, plane)
R = su.rot3d(plane[:3],n_front)
@@ -248,7 +248,7 @@ def get_text_placement_mask(xyz,mask,plane,pad=2,viz=False):
# the same scale as the target region:
s = rescale_frontoparallel(pts_tmp,boxR,pts[0])
boxR *= s
- for i in xrange(len(pts_fp)):
+ for i in range(len(pts_fp)):
pts_fp[i] = s*((pts_fp[i]-mu[None,:]).dot(R2d.T) + mu[None,:])
# paint the unrotated contour points:
@@ -258,7 +258,7 @@ def get_text_placement_mask(xyz,mask,plane,pad=2,viz=False):
place_mask = 255*np.ones((int(np.ceil(COL))+pad, int(np.ceil(ROW))+pad), 'uint8')
- pts_fp_i32 = [(pts_fp[i]+minxy[None,:]).astype('int32') for i in xrange(len(pts_fp))]
+ pts_fp_i32 = [(pts_fp[i]+minxy[None,:]).astype('int32') for i in range(len(pts_fp))]
cv2.drawContours(place_mask,pts_fp_i32,-1,0,
thickness=cv2.FILLED,
lineType=8,hierarchy=hier)
@@ -279,8 +279,7 @@ def get_text_placement_mask(xyz,mask,plane,pad=2,viz=False):
plt.imshow(mask)
plt.subplot(1,2,2)
plt.imshow(~place_mask)
- plt.hold(True)
- for i in xrange(len(pts_fp_i32)):
+ for i in range(len(pts_fp_i32)):
plt.scatter(pts_fp_i32[i][:,0],pts_fp_i32[i][:,1],
edgecolors='none',facecolor='g',alpha=0.5)
plt.show()
@@ -318,7 +317,7 @@ def mean_seg(rgb,seg,label):
plt.close(fignum)
plt.figure(fignum)
ims = [rgb,mim,depth,img]
- for i in xrange(len(ims)):
+ for i in range(len(ims)):
plt.subplot(2,2,i+1)
plt.imshow(ims[i])
plt.show(block=False)
@@ -337,9 +336,9 @@ def viz_regions(img,xyz,seg,planes,labels):
xyz_region = xyz[mask,:]
su.visualize_plane(xyz_region,np.array(planes[i]))
- mym.view(180,180)
- mym.orientation_axes()
- mym.show(True)
+# mym.view(180,180)
+# mym.orientation_axes()
+# mym.show(True)
def viz_textbb(fignum,text_im, bb_list,alpha=1.0):
"""
@@ -349,12 +348,11 @@ def viz_textbb(fignum,text_im, bb_list,alpha=1.0):
plt.close(fignum)
plt.figure(fignum)
plt.imshow(text_im)
- plt.hold(True)
H,W = text_im.shape[:2]
- for i in xrange(len(bb_list)):
+ for i in range(len(bb_list)):
bbs = bb_list[i]
ni = bbs.shape[-1]
- for j in xrange(ni):
+ for j in range(ni):
bb = bbs[:,:,j]
bb = np.c_[bb,bb[:,0]]
plt.plot(bb[0,:], bb[1,:], 'r', linewidth=2, alpha=alpha)
@@ -552,7 +550,7 @@ def char2wordBB(self, charBB, text):
bb_idx = np.r_[0, np.cumsum([len(w) for w in wrds])]
wordBB = np.zeros((2,4,len(wrds)), 'float32')
- for i in xrange(len(wrds)):
+ for i in range(len(wrds)):
cc = charBB[:,:,bb_idx[i]:bb_idx[i+1]]
# fit a rotated-rectangle:
@@ -570,7 +568,7 @@ def char2wordBB(self, charBB, text):
cc[3,:]].T
perm4 = np.array(list(itertools.permutations(np.arange(4))))
dists = []
- for pidx in xrange(perm4.shape[0]):
+ for pidx in range(perm4.shape[0]):
d = np.sum(np.linalg.norm(box[perm4[pidx],:]-cc_tblr,axis=1))
dists.append(d)
wordBB[:,:,i] = box[perm4[np.argmin(dists)],:].T
@@ -626,10 +624,10 @@ def render_text(self,rgb,depth,seg,area,label,ninstance=1,viz=False):
return []
res = []
- for i in xrange(ninstance):
+ for i in range(ninstance):
place_masks = copy.deepcopy(regions['place_mask'])
- print colorize(Color.CYAN, " ** instance # : %d"%i)
+ print (" ** instance # : %d"%i)
idict = {'img':[], 'charBB':None, 'wordBB':None, 'txt':None}
@@ -649,24 +647,10 @@ def render_text(self,rgb,depth,seg,area,label,ninstance=1,viz=False):
reg_range = np.arange(NUM_REP * num_txt_regions) % num_txt_regions
for idx in reg_range:
ireg = reg_idx[idx]
- try:
- if self.max_time is None:
- txt_render_res = self.place_text(img,place_masks[ireg],
- regions['homography'][ireg],
- regions['homography_inv'][ireg])
- else:
- with time_limit(self.max_time):
- txt_render_res = self.place_text(img,place_masks[ireg],
- regions['homography'][ireg],
- regions['homography_inv'][ireg])
- except TimeoutException, msg:
- print msg
- continue
- except:
- traceback.print_exc()
- # some error in placing text on the region
- continue
-
+ txt_render_res = self.place_text(img,place_masks[ireg],
+ regions['homography'][ireg],
+ regions['homography_inv'][ireg])
+
if txt_render_res is not None:
placed = True
img,text,bb,collision_mask = txt_render_res
@@ -687,6 +671,6 @@ def render_text(self,rgb,depth,seg,area,label,ninstance=1,viz=False):
viz_textbb(1,img, [idict['wordBB']], alpha=1.0)
viz_masks(2,img,seg,depth,regions['label'])
# viz_regions(rgb.copy(),xyz,seg,regions['coeff'],regions['label'])
- if i < ninstance-1:
- raw_input(colorize(Color.BLUE,'continue?',True))
+ # if i < ninstance-1:
+ # input('continue?')
return res
diff --git a/text_utils.py b/text_utils.py
index be6080b..2bc0e56 100644
--- a/text_utils.py
+++ b/text_utils.py
@@ -8,7 +8,8 @@
import os.path as osp
import random, os
import cv2
-import cPickle as cp
+
+import _pickle as cp
import scipy.signal as ssig
import scipy.stats as sstat
import pygame, pygame.locals
@@ -17,7 +18,6 @@
from PIL import Image
import math
from common import *
-import codecs
from logger import logger
import nltk, re, pprint
@@ -27,10 +27,12 @@
from nltk.text import Text
from nltk.corpus.reader.chasen import *
import subprocess
+import pickle
def sample_weighted(p_dict):
- ps = p_dict.keys()
- return p_dict[np.random.choice(ps,p=ps)]
+ key_list = list(p_dict.keys())
+ chosen = key_list[np.random.randint(0, len(key_list))]
+ return p_dict[chosen]
def move_bb(bbs, t):
"""
@@ -57,7 +59,7 @@ def crop_safe(arr, rect, bbs=[], pad=0):
v1 = [min(arr.shape[0], rect[0]+rect[2]), min(arr.shape[1], rect[1]+rect[3])]
arr = arr[v0[0]:v1[0],v0[1]:v1[1],...]
if len(bbs) > 0:
- for i in xrange(len(bbs)):
+ for i in range(len(bbs)):
bbs[i,0] -= v0[0]
bbs[i,1] -= v0[1]
return arr, bbs
@@ -195,9 +197,9 @@ def render_curved(self, font, word_text):
# baseline state
mid_idx = wl//2
BS = self.baselinestate.get_sample()
- curve = [BS['curve'](i-mid_idx) for i in xrange(wl)]
+ curve = [BS['curve'](i-mid_idx) for i in range(wl)]
curve[mid_idx] = -np.sum(curve) / (wl-1)
- rots = [-int(math.degrees(math.atan(BS['diff'](i-mid_idx)/(font.size/2)))) for i in xrange(wl)]
+ rots = [-int(math.degrees(math.atan(BS['diff'](i-mid_idx)/(font.size/2)))) for i in range(wl)]
bbs = []
# place middle char
@@ -213,7 +215,7 @@ def render_curved(self, font, word_text):
# render chars to the left and right:
last_rect = rect
ch_idx = []
- for i in xrange(wl):
+ for i in range(wl):
#skip the middle character
if i==mid_idx:
bbs.append(mid_ch_bb)
@@ -325,7 +327,7 @@ def bb_xywh2coords(self,bbs):
"""
n,_ = bbs.shape
coords = np.zeros((2,4,n))
- for i in xrange(n):
+ for i in range(n):
coords[:,:,i] = bbs[i,:2][:,None]
coords[0,1,i] += bbs[i,2]
coords[:,2,i] += bbs[i,2:4]
@@ -431,13 +433,21 @@ def __init__(self, data_dir='data'):
font_model_path = osp.join(data_dir, 'models/font_px2pt.cp')
# get character-frequencies in the English language:
- with open(char_freq_path,'r') as f:
- self.char_freq = cp.load(f)
+ with open(char_freq_path,'rb') as f:
+ #self.char_freq = cp.load(f)
+ u = pickle._Unpickler(f)
+ u.encoding = 'latin1'
+ p = u.load()
+ self.char_freq = p
# get the model to convert from pixel to font pt size:
- with open(font_model_path,'r') as f:
- self.font_model = cp.load(f)
-
+ with open(font_model_path,'rb') as f:
+ #self.font_model = cp.load(f)
+ u = pickle._Unpickler(f)
+ u.encoding = 'latin1'
+ p = u.load()
+ self.font_model = p
+
# get the names of fonts to use:
self.FONT_LIST = osp.join(data_dir, 'fonts/fontlist.txt')
self.fonts = [os.path.join(data_dir,'fonts',f.strip()) for f in open(self.FONT_LIST)]
@@ -455,7 +465,7 @@ def get_aspect_ratio(self, font, size=None):
# get the [height,width] of each character:
try:
sizes = font.get_metrics(chars,size)
- good_idx = [i for i in xrange(len(sizes)) if sizes[i] is not None]
+ good_idx = [i for i in range(len(sizes)) if sizes[i] is not None]
sizes,w = [sizes[i] for i in good_idx], w[good_idx]
sizes = np.array(sizes).astype('float')[:,[3,4]]
r = np.abs(sizes[:,1]/sizes[:,0]) # width/height
@@ -558,11 +568,11 @@ def __init__(self, min_nchar, fn, lang="ENG"):
# convert fs into chasen file
_, ext = os.path.splitext(os.path.basename(fn))
fn_chasen = fn.replace(ext, ".chasen")
- print "Convert {} into {}".format(fn, fn_chasen)
+ print ("Convert {} into {}".format(fn, fn_chasen))
cmd = "mecab -Ochasen {} > {}".format(fn, fn_chasen)
- print "The following cmd below was executed to convert into chasen (for Japanese)"
- print "\t{}".format(cmd)
+ print ("The following cmd below was executed to convert into chasen (for Japanese)")
+ print ("\t{}".format(cmd))
p = subprocess.call(cmd, shell=True)
data = ChasenCorpusReader('./', fn_chasen, encoding='utf-8')
@@ -636,7 +646,7 @@ def center_align(self, lines):
"""
ls = [len(l) for l in lines]
max_l = max(ls)
- for i in xrange(len(lines)):
+ for i in range(len(lines)):
l = lines[i].strip()
dl = max_l-ls[i]
lspace = dl//2
@@ -676,9 +686,9 @@ def h_lines(niter=100):
lines[i] = lines[i][:len(lines[i])-lines[i][::-1].find(' ')].strip()
if not np.all(self.is_good(lines,f)):
- return #None
+ return
else:
- return lines
+ return [''.join(line) if type(line)==list else line for line in lines]
def sample(self, nline_max,nchar_max,kind='WORD'):
return self.fdict[kind](nline_max,nchar_max)
@@ -708,7 +718,7 @@ def sample_line(self,nline_max,nchar_max):
# get number of words:
nword = [self.p_line_nword[2]*sstat.beta.rvs(a=self.p_line_nword[0], b=self.p_line_nword[1])
- for _ in xrange(nline)]
+ for _ in range(nline)]
nword = [max(1,int(np.ceil(n))) for n in nword]
lines = self.get_lines(nline, nword, nchar_max, f=0.35)
@@ -724,7 +734,7 @@ def sample_para(self,nline_max,nchar_max):
# get number of words:
nword = [self.p_para_nword[2]*sstat.beta.rvs(a=self.p_para_nword[0], b=self.p_para_nword[1])
- for _ in xrange(nline)]
+ for _ in range(nline)]
nword = [max(1,int(np.ceil(n))) for n in nword]
lines = self.get_lines(nline, nword, nchar_max, f=0.35)
diff --git a/visualize_results.py b/visualize_results.py
index 92091ef..c143635 100644
--- a/visualize_results.py
+++ b/visualize_results.py
@@ -24,26 +24,25 @@ def viz_textbb(text_im, charBB_list, wordBB, alpha=1.0):
plt.close(1)
plt.figure(1)
plt.imshow(text_im)
- plt.hold(True)
H,W = text_im.shape[:2]
# plot the character-BB:
- for i in xrange(len(charBB_list)):
+ for i in range(len(charBB_list)):
bbs = charBB_list[i]
ni = bbs.shape[-1]
- for j in xrange(ni):
+ for j in range(ni):
bb = bbs[:,:,j]
bb = np.c_[bb,bb[:,0]]
plt.plot(bb[0,:], bb[1,:], 'r', alpha=alpha/2)
# plot the word-BB:
- for i in xrange(wordBB.shape[-1]):
+ for i in range(wordBB.shape[-1]):
bb = wordBB[:,:,i]
bb = np.c_[bb,bb[:,0]]
plt.plot(bb[0,:], bb[1,:], 'g', alpha=alpha)
# visualize the indiv vertices:
vcol = ['r','g','b','k']
- for j in xrange(4):
+ for j in range(4):
plt.scatter(bb[0,j],bb[1,j],color=vcol[j])
plt.gca().set_xlim([0,W-1])
@@ -53,7 +52,7 @@ def viz_textbb(text_im, charBB_list, wordBB, alpha=1.0):
def main(db_fname):
db = h5py.File(db_fname, 'r')
dsets = sorted(db['data'].keys())
- print "total number of images : ", colorize(Color.RED, len(dsets), highlight=True)
+ print ("total number of images : ", colorize(Color.RED, len(dsets), highlight=True))
for k in dsets:
rgb = db['data'][k][...]
charBB = db['data'][k].attrs['charBB']
@@ -61,12 +60,12 @@ def main(db_fname):
txt = db['data'][k].attrs['txt']
viz_textbb(rgb, [charBB], wordBB)
- print "image name : ", colorize(Color.RED, k, bold=True)
- print " ** no. of chars : ", colorize(Color.YELLOW, charBB.shape[-1])
- print " ** no. of words : ", colorize(Color.YELLOW, wordBB.shape[-1])
- print " ** text : ", colorize(Color.GREEN, txt)
+ print ("image name : ", colorize(Color.RED, k, bold=True))
+ print (" ** no. of chars : ", colorize(Color.YELLOW, charBB.shape[-1]))
+ print (" ** no. of words : ", colorize(Color.YELLOW, wordBB.shape[-1]))
+ print (" ** text : ", colorize(Color.GREEN, txt))
- if 'q' in raw_input("next? ('q' to exit) : "):
+ if 'q' in input("next? ('q' to exit) : "):
break
db.close()