def gen_ds_list(listfile=dslist): dataset = list() tags_to_labels, labels_to_tags = gen_ydata.load_mappings() with open(listfile, 'r') as fd: for line in fd: [sample_dir_path, s_n_imgs] = line.strip().split('\t') n_imgs = int(s_n_imgs) m = matcher.match(sample_dir_path) if (m): g = m.groups() matching_classes = decode_classes(g) y = classes_to_y(matching_classes) fullfpath = "%s/%s" % (dsdir, sample_dir_path) csvpath = "%s.csv" % fullfpath if (os.path.exists(csvpath)): #Fully load CSV data csvdata = gen_ydata.parse_csv(csvpath, n_imgs, mapping=tags_to_labels) else: csvdata = None dataset.append( (fullfpath, n_imgs, y, matching_classes, csvdata)) else: print("WARNING: dir %s does not match pattern" % fpath) return dataset
import cv2 import classif_mehdi_fcn import analytics import gen_ydata from threading import Thread, Lock #from Queue import Queue from skimage import io # from socketIO_client_nexus import SocketIO, LoggingNamespace # #socketIO = SocketIO('93.24.79.14', 3333) # #socketIO = SocketIO('localhost', 3333, LoggingNamespace) # tags_to_labels, labels_to_tags = gen_ydata.load_mappings() def entropy_from_hmaps(hmaps): entropy = np.zeros(hmaps.shape[:2]) for i in range(len(classif_mehdi_fcn.classes)): entropy += - hmaps[:,:,i]*np.log(hmaps[:,:,i]) return entropy def compute_histograms(hmaps,entropy_th=None): histos = list() if(entropy_th): entropy = entropy_from_hmaps(hmaps) good_pixels = np.array(entropy<entropy_th,dtype=np.float)