test_inds[value_inds[n:]] = True return np.where(train_inds)[0], np.where(test_inds)[0] #============================================================================== # SET THESE PARAMETERS! #============================================================================== curdb_dir = 'roi' data_dir = r'c:\Users\picturio\OneDrive\KS-XR\Images' #============================================================================== # RUN CONFIG #============================================================================== cfg = train_params(data_dir, curdb_dir=curdb_dir, training_id=training_id) """ Read data description file """ df_db = pd.read_csv(cfg.db_file, delimiter=';') """ Select classes to process """ df_filtered = df_db.copy() df_labeled = df_db[['Filename']].copy() df_labeled['category'] = df_db[['Class name']] df_labeled.columns = ['image', 'category'] """
# SET THESE PARAMETERS! #============================================================================== training_id='20180308' curdb_dir='db_cropped_rot' #data_dir=os.path.join('C:','Users','picturio','OneDrive','WaterScope') #data_dir=os.path.join('E:','OneDrive','WaterScope') data_dir=os.path.join('/','home','mikesz','ownCloud','WaterScope') # cropped results are saved here #save_dir=os.path.join('D:','DATA','WaterScope','tmp_problem') save_dir=os.path.join('/','home','mikesz','Data','WaterScope','tmp_cropped') #============================================================================== # RUN CONFIG #============================================================================== cfg=train_params(data_dir,base_db='db_categorized',curdb_dir=curdb_dir,training_id=training_id) #============================================================================== # Create database using folder names #============================================================================== image_list=fh.imagelist_in_depth(cfg.base_imagedb_dir,level=1) """ Class names from folder names """ class_names=[os.path.dirname(f).split(os.sep)[-1] for f in image_list] df_db = pd.DataFrame(data={'Filename':image_list,'Class name':class_names}) #df_db=df_db[df_db['Class name']=='Others'] #==============================================================================
numFeature = image_height * image_width * num_channels #============================================================================== # SET training parameters #============================================================================== max_epochs = 150 model_func = create_shallow_model epoch_size = 3300 # training minibatch_size = 128 # training #============================================================================== # RUN CONFIG #============================================================================== cfg = train_params(data_dir, training_id=training_id) data_mean_file = os.path.join(cfg.train_dir, 'data_mean.xml') model_file = os.path.join(cfg.train_dir, 'cnn_model.dnn') model_temp_file = os.path.join(cfg.train_dir, 'cnn_model_temp.dnn') train_log_file = os.path.join(cfg.train_log_dir, 'progress_log.txt') train_map_image = os.path.join(cfg.train_dir, 'train_map_image.txt') test_map_image = os.path.join(cfg.train_dir, 'test_map_image.txt') # # Evaluation action # def evaluate_test(input_map, reader_test,
import matplotlib.pyplot as plt #from cntk.device import try_set_default_device, gpu from cntk import cross_entropy_with_softmax, classification_error, input_variable, softmax, element_times from cntk.io import MinibatchSource, ImageDeserializer, StreamDef, StreamDefs, transforms from cntk import Trainer, UnitType from cntk import momentum_sgd, learning_rate_schedule, momentum_as_time_constant_schedule #from cntk.learners import momentum_schedule from cntk.logging import log_number_of_parameters, ProgressPrinter, TensorBoardProgressWriter from src_train.model_functions import create_basic_model, create_advanced_model from src_train.train_config import train_params #from src_train.readers import create_reader data_dir = os.path.join(r'C:\Users', 'picturio', 'OneDrive\WaterScope') cfg = train_params(data_dir, crop=True, training_id='20171113') model_file = os.path.join(cfg.train_dir, 'cnn_model.dnn') model_temp_file = os.path.join(cfg.train_dir, 'cnn_model_temp.dnn') train_log_file = os.path.join(cfg.train_log_dir, 'progress_log.txt') train_map = os.path.join(cfg.train_dir, 'train_map.txt') test_map = os.path.join(cfg.train_dir, 'test_map.txt') # GET train and test map from prepare4train data_mean_file = os.path.join(cfg.train_dir, 'data_mean.xml') # model dimensions image_height = 64 image_width = 64
from src_train.train_config import train_params import src_tools.file_helper as fh from matplotlib import pyplot as plt import matplotlib.patches as patches import crop #data_dir=os.path.join('C:','Users','picturio','OneDrive','WaterScope') #data_dir=os.path.join('E:','OneDrive','WaterScope') data_dir=os.path.join('/','home','mikesz','ownCloud','WaterScope') cfg=train_params(data_dir,base_db='db_categorized',curdb_dir='crop_problems') #save_dir=os.path.join('D:\\','DATA','WaterScope','tmp_problem') save_dir=os.path.join('/','home','mikesz','Data','WaterScope','tmp_problem') #image_list=fh.imagelist_in_depth(cfg.base_imagedb_dir,level=2) image_list=fh.imagelist_in_depth(cfg.curdb_dir,level=2) #for image_file in image_list: # img = Image.open(image_file) # img_square=crop.crop(img,pad_rate=0.25, # save_file=os.path.join(data_dir,'Images','tmp',os.path.basename(image_file)), # category='dummy')