def get_imdb(dataset, splitName):
    imdbPaths = getImdbPaths(dataset, splitName)
    imdb = coco_voc.coco_voc(dataset,
                             splitName,
                             image_path=imdbPaths['imageDir'],
                             captionJsonPath=imdbPaths['jsonPath'])
    return imdb
import os
import sg_utils as utils
import coco_voc
import shutil

# Make directories
for i in xrange(60):
  utils.mkdir_if_missing(os.path.join('..', 'data', 'images', '{:02d}'.format(i)))

# Copy files over
sets = ['train', 'val', 'test']
for set_ in sets:
  imdb = coco_voc.coco_voc(set_)
  for i in xrange(imdb.num_images):
    in_file = os.path.join('../data', set_ + '2014', \
      'COCO_{}2014_{:012d}.jpg'.format(set_, imdb.image_index[i])); 
    out_file = imdb.image_path_at(i)
    # print in_file, out_file
    shutil.copyfile(in_file, out_file)
    utils.tic_toc_print(1, ' Copying images [{}]: {:06d} / {:06d}\n'.format(set_, i, imdb.num_images));
Beispiel #3
0
import os
import sg_utils as utils
import coco_voc
import shutil

# Make directories
for i in xrange(60):
  utils.mkdir_if_missing(os.path.join('data', 'images', '{:02d}'.format(i)))

# Copy files over
sets = ['train', 'val', 'test']
for set_ in sets:
  imdb = coco_voc.coco_voc(set_)
  for i in xrange(imdb.num_images):
    in_file = os.path.join(set_ + '2014', \
      'COCO_{}2014_{:012d}.jpg'.format(set_, imdb.image_index[i])); 
    out_file = imdb.image_path_at(i)
    # print in_file, out_file
    shutil.copyfile(in_file, out_file)
    utils.tic_toc_print(1, ' Copying images [{}]: {:06d} / {:06d}\n'.format(set_, i, imdb.num_images));
Beispiel #4
0
        type=str)

    if len(sys.argv) == 1:
        parser.print_help()
        sys.exit(1)

    args = parser.parse_args()
    return args

if __name__ == '__main__':
    args = parse_args()

    print('Called with args:')
    print(args)

    imdb = coco_voc.coco_voc('test')
    vocab = utils.load_variables(args.vocab_file)
    gt_label = preprocess.get_vocab_counts(
        imdb.image_index,
        imdb.coco_caption_data,
        5,
        vocab
        )
    det_file = args.det_file
    det_dir = os.path.dirname(det_file) # get root dir of det_file

    eval_file = os.path.join(det_dir, imdb.name + '_eval.pkl')
    benchmark(imdb, vocab, gt_label, 5, det_file, eval_file=eval_file)

    map_file = args.map_file
    gt_label_det = preprocess.get_vocab_counts_det(
Beispiel #5
0
  # # set up caffe
  caffe.set_mode_gpu()
  if args.gpu_id is not None:
    caffe.set_device(args.gpu_id)
 
  # Load the vocabulary
  vocab = utils.load_variables(args.vocab_file)
  
  if args.task == 'compute_targets':
    
    imdb = []
    output_dir = args.train_dir
    sets = ['train', 'val']
    for i, imset in enumerate([args.train_set, args.val_set]):
      imdb.append(coco_voc.coco_voc(imset))
      print 'Loaded dataset {:s}'.format(imdb[i].name)
      
      # Compute targets for the file
      counts = preprocess.get_vocab_counts(imdb[i].image_index, \
          imdb[i].coco_caption_data, 5, vocab)
      
      if args.write_labels:
        label_file = os.path.join(output_dir, 'labels_' + sets[i] + '.h5') 
        print 'Writing labels to {}'.format(label_file)
        with h5py.File(label_file, 'w') as f:
          for j in xrange(imdb[i].num_images):
            ind = imdb[i].image_index[j]
            ind_str = '{:02d}/{:d}'.format(int(math.floor(ind)/1e4), ind)
            l = f.create_dataset('/labels-{}'.format(ind_str), (1, 1, counts.shape[1], 1), dtype = 'f')
            c = counts[j,:].copy(); c = c > 0; c = c.astype(np.float32); c = c.reshape((1, 1, c.size, 1))
def get_imdb(dataset, splitName):
  imdbPaths = getImdbPaths(dataset, splitName);
  imdb = coco_voc.coco_voc(dataset, splitName, image_path=imdbPaths['imageDir'], captionJsonPath = imdbPaths['jsonPath'])
  return imdb;