def main(): print('Loading APC2015berkeley...') dataset = load_APC2015berkeley() N = len(dataset.target) print(''' N: {0} dataset: {1}'''.format(N, dataset)) bg_label = dataset.target_names.index('__background__') n_labels = len(dataset.target_names) max_batch_size = 10 for i in xrange(0, N, max_batch_size): t_start = time.time() fname_batch = dataset.filenames[i:i+max_batch_size] label_batch = dataset.target[i:i+max_batch_size] blob_batch, bbox_batch, label_batch, roi_delta_batch =\ load_batch_APC2015berkeley(fname_batch, label_batch, bg_label, n_labels) # show stats elapsed_time = time.time() - t_start head_fname_batch = ['/'.join(f.split('/')[-2:]) for f in fname_batch] print(''' elapsed_time: {0} [s] fnames: {1} blob: {2} bboxes: {3} label: {4} roi_delta: {5}'''.format(elapsed_time, head_fname_batch, blob_batch.shape, bbox_batch.shape, label_batch.shape, roi_delta_batch.shape))
def main(): dataset = load_APC2015berkeley() n_class = len(dataset.target_names) model = load_pretrained_model( n_class=n_class, bg_label=dataset.background_label) model.to_gpu() trainer = FastRCNNTrainer(model, dataset=dataset) trainer.main_loop(batch_size=3, epoch_size=100)