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)
