Windows support and CMake for lanms

.gitignore

@@ -105,3 +105,16 @@ ENV/
 
 # result files for demo
 static/results
+
+# model files
+models/
+
+# binary images
+*.jpg
+*.png
+
+# output directory
+outputs/
+
+# training dataset directory
+data/

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "lanms/pybind11"]
+	path = lanms/pybind11
+	url = https://github.com/pybind/pybind11/

eval.py

@@ -8,15 +8,10 @@
 import locality_aware_nms as nms_locality
 import lanms
 
-tf.app.flags.DEFINE_string('test_data_path', '/tmp/ch4_test_images/images/', '')
-tf.app.flags.DEFINE_string('gpu_list', '0', '')
-tf.app.flags.DEFINE_string('checkpoint_path', '/tmp/east_icdar2015_resnet_v1_50_rbox/', '')
-tf.app.flags.DEFINE_string('output_dir', '/tmp/ch4_test_images/images/', '')
-tf.app.flags.DEFINE_bool('no_write_images', False, 'do not write images')
-
 import model
 from icdar import restore_rectangle
 
+import flags
 FLAGS = tf.app.flags.FLAGS
 
 def get_images():
@@ -68,15 +63,15 @@ def resize_image(im, max_side_len=2400):
     return im, (ratio_h, ratio_w)
 
 
-def detect(score_map, geo_map, timer, score_map_thresh=0.8, box_thresh=0.1, nms_thres=0.2):
+def detect(score_map, geo_map, timer, score_map_thresh=0.8, box_thresh=0.1, nms_thresh=0.2):
     '''
     restore text boxes from score map and geo map
     :param score_map:
     :param geo_map:
     :param timer:
-    :param score_map_thresh: threshhold for score map
-    :param box_thresh: threshhold for boxes
-    :param nms_thres: threshold for nms
+    :param score_map_thresh: threshold for score map
+    :param box_thresh: threshold for boxes
+    :param nms_thresh: threshold for nms
     :return:
     '''
     if len(score_map.shape) == 4:
@@ -96,8 +91,8 @@ def detect(score_map, geo_map, timer, score_map_thresh=0.8, box_thresh=0.1, nms_
     timer['restore'] = time.time() - start
     # nms part
     start = time.time()
-    # boxes = nms_locality.nms_locality(boxes.astype(np.float64), nms_thres)
-    boxes = lanms.merge_quadrangle_n9(boxes.astype('float32'), nms_thres)
+    # boxes = nms_locality.nms_locality(boxes.astype(np.float64), nms_thresh)
+    boxes = lanms.merge_quadrangle_n9(boxes.astype('float32'), nms_thresh)
     timer['nms'] = time.time() - start
 
     if boxes.shape[0] == 0:

flags.py

@@ -0,0 +1,41 @@
+import tensorflow as tf
+
+# required for both training and testing
+tf.app.flags.DEFINE_string('checkpoint_path', './models/east_icdar2015_resnet_v1_50_rbox',
+                           'model folder that contains checkpoint, index and meta')
+tf.app.flags.DEFINE_integer('text_scale', 512, '')
+
+# testing parameters
+tf.app.flags.DEFINE_string('test_data_path', './training_samples',
+                           'folder that contains images to test')
+tf.app.flags.DEFINE_string('output_dir', './outputs',
+                           'result will be written to this folder')
+tf.app.flags.DEFINE_bool('no_write_images', False,
+                         'do not write images')
+
+# training parameters
+tf.app.flags.DEFINE_boolean('restore', False,
+                            'whether to restore from checkpoint')
+tf.app.flags.DEFINE_integer('max_steps', 100000, '')
+tf.app.flags.DEFINE_integer('save_checkpoint_steps', 1000, '')
+tf.app.flags.DEFINE_integer('save_summary_steps', 100, '')
+tf.app.flags.DEFINE_integer('batch_size_per_gpu', 14, '')
+tf.app.flags.DEFINE_integer('num_readers', 16, '')
+tf.app.flags.DEFINE_string('training_data_path', './training_samples/',
+                           'training dataset to use')
+tf.app.flags.DEFINE_string('pretrained_model_path', './models/resnet_v1_50.ckpt', '')
+tf.app.flags.DEFINE_string('gpu_list', '0', '')
+tf.app.flags.DEFINE_integer('input_size', 512, '')
+tf.app.flags.DEFINE_float('learning_rate', 0.0001, '')
+tf.app.flags.DEFINE_float('moving_average_decay', 0.997, '')
+tf.app.flags.DEFINE_integer('max_image_large_side', 1280, 'max image size of training')
+tf.app.flags.DEFINE_integer('max_text_size', 800,
+                            'if the text in the input image is bigger than this, '
+                            'then we resize the image according to this')
+tf.app.flags.DEFINE_integer('min_text_size', 10,
+                            'if the text size is smaller than this, we ignore it during training')
+tf.app.flags.DEFINE_float('min_crop_side_ratio', 0.1,
+                          'when doing random crop from input image, '
+                          'the min length of min(H, W)')
+tf.app.flags.DEFINE_string('geometry', 'RBOX',
+                           'which geometry to generate, RBOX or QUAD')

icdar.py

@@ -10,26 +10,10 @@
 import matplotlib.patches as Patches
 from shapely.geometry import Polygon
 
-import tensorflow as tf
-
 from data_util import GeneratorEnqueuer
 
-tf.app.flags.DEFINE_string('training_data_path', '/data/ocr/icdar2015/',
-                           'training dataset to use')
-tf.app.flags.DEFINE_integer('max_image_large_side', 1280,
-                            'max image size of training')
-tf.app.flags.DEFINE_integer('max_text_size', 800,
-                            'if the text in the input image is bigger than this, then we resize'
-                            'the image according to this')
-tf.app.flags.DEFINE_integer('min_text_size', 10,
-                            'if the text size is smaller than this, we ignore it during training')
-tf.app.flags.DEFINE_float('min_crop_side_ratio', 0.1,
-                          'when doing random crop from input image, the'
-                          'min length of min(H, W')
-tf.app.flags.DEFINE_string('geometry', 'RBOX',
-                           'which geometry to generate, RBOX or QUAD')
-
-
+import flags
+import tensorflow as tf
 FLAGS = tf.app.flags.FLAGS
 
 

lanms/.gitignore

@@ -1 +1,2 @@
 adaptor.so
+build

lanms/CMakeLists.txt

@@ -0,0 +1,26 @@
+cmake_minimum_required(VERSION 3.4)
+project(lanms)
+
+set(CMAKE_CXX_STANDARD 11)
+if(WIN32)
+  set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
+else(WIN32)
+  add_compile_options(-Wall -Wextra -Wpedantic)
+  add_compile_options(-pthread)
+endif(WIN32)
+
+set(SOURCES "lanms.cpp" "include/clipper/clipper.cpp")
+
+add_library(lanms_library SHARED ${SOURCES})
+target_link_libraries(lanms_library)
+target_include_directories(lanms_library PUBLIC ${CMAKE_CURRENT_LIST_DIR})
+
+include_directories("include")
+
+add_executable(main main.cpp ${SOURCES})
+target_link_libraries(main)
+
+add_subdirectory("pybind11")
+pybind11_add_module(adaptor "adaptor.cpp" ${SOURCES})
+target_link_libraries(adaptor PRIVATE lanms_library)
+target_compile_definitions(adaptor PRIVATE)

lanms/__init__.py

@@ -1,13 +1,5 @@
-import subprocess
-import os
 import numpy as np
 
-BASE_DIR = os.path.dirname(os.path.realpath(__file__))
-
-if subprocess.call(['make', '-C', BASE_DIR]) != 0:  # return value
-    raise RuntimeError('Cannot compile lanms: {}'.format(BASE_DIR))
-
-
 def merge_quadrangle_n9(polys, thres=0.3, precision=10000):
     from .adaptor import merge_quadrangle_n9 as nms_impl
     if len(polys) == 0:
@@ -17,4 +9,3 @@ def merge_quadrangle_n9(polys, thres=0.3, precision=10000):
     ret = np.array(nms_impl(p, thres), dtype='float32')
     ret[:,:8] /= precision
     return ret
-