def extract_daquar_feature(): image_dir = '/data/lisatmp4/taesup/data/daquar/image' train_list_filename = '/data/lisatmp4/taesup/data/daquar/train_image_list.txt' test_list_filename = '/data/lisatmp4/taesup/data/daquar/test_image_list.txt' train_desc_filename = '/data/lisatmp4/taesup/data/daquar/train_desc.pkl' test_desc_filename = '/data/lisatmp4/taesup/data/daquar/test_desc.pkl' with open(train_list_filename, 'r') as fp: train_images = fp.readlines() train_features = OrderedDict() for i, train_image in enumerate(train_images): image_name = train_image.split()[0] image_path = image_dir + '/' + image_name + '.png' image_data = load_resize_image(image_path)[None,] features = vgg_feature_extractor(image_data.transpose((0, 3, 1, 2)))[0][0] sparse_features = sparse.csr_matrix(features.reshape((features.shape[0],features.shape[1]*features.shape[2]))) train_features[image_name] = sparse_features print 'daquar train {}th image done'.format(i) with open(train_desc_filename,'wb') as fp: pkl.dump(train_features,fp) with open(test_list_filename, 'r') as fp: test_images = fp.readlines() test_features = OrderedDict() for i, test_image in enumerate(test_images): image_name = test_image.split()[0] image_path = image_dir + '/' + image_name + '.png' image_data = load_resize_image(image_path)[None,] features = vgg_feature_extractor(image_data.transpose((0, 3, 1, 2)))[0][0] sparse_features = sparse.csr_matrix(features.reshape((features.shape[0],features.shape[1]*features.shape[2]))) test_features[image_name] = sparse_features print 'daquar test {}th image done'.format(i) with open(test_desc_filename,'wb') as fp: pkl.dump(test_features,fp)
def extract_coco_feature(): image_dirs = ['/data/lisatmp4/taesup/data/coco/images/train2014', '/data/lisatmp4/taesup/data/coco/images/val2014', '/data/lisatmp4/taesup/data/coco/images/test2014', '/data/lisatmp4/taesup/data/coco/images/test2015'] desc_files = ['/data/lisatmp4/taesup/data/coco/descriptors/train2014_desc.pkl', '/data/lisatmp4/taesup/data/coco/descriptors/val2014_desc.pkl', '/data/lisatmp4/taesup/data/coco/descriptors/test2014_desc.pkl', '/data/lisatmp4/taesup/data/coco/descriptors/test2015_desc.pkl'] for image_dir, desc_file in zip(image_dirs, desc_files): image_list = glob.glob(image_dir+'/*') feature_dict = OrderedDict() for image_path in image_list: image_data = load_resize_image(image_path)[None,] features = vgg_feature_extractor(image_data.transpose((0, 3, 1, 2)))[0][0] sparse_features = sparse.csr_matrix(features.reshape((features.shape[0],features.shape[1]*features.shape[2]))) image_name = image_path.replace(image_dir,'') image_name = image_name.split('_')[2][:-4] image_name = int(image_name) feature_dict[image_name] = sparse_features with open(desc_file,'wb') as fp: pkl.dump(feature_dict,fp)