plt.plot(plot_dict['x_valid'], plot_dict['y_sim'])
plt.ylabel('sim metric')
plt.subplot(4, 1, 4)
plt.plot(plot_dict['x_valid'], plot_dict['y_kld'])
plt.xlabel('iter')
plt.ylabel('kld metric')
_step = 0
while _step < max_iter:
# if connection:
# frame_data, density_data, reference_density = tranining_dataset.get_frame_connection_c3d(mini_batch=batch, phase='training', density_length='full', data_augmentation=data_augmentation)
# else:
# frame_data, density_data = tranining_dataset.get_frame_c3d(mini_batch=batch, phase='training', density_length='full', data_augmentation=data_augmentation)
frame_data, density_data = tranining_dataset.get_frame_c3d(
mini_batch=batch,
phase='training',
density_length='full',
data_augmentation=data_augmentation)
solver.net.blobs['data'].data[...] = frame_data
solver.net.blobs['gt'].data[...] = density_data
# if connection:
# solver.net.blobs['reference_density'].data[...] = reference_density
solver.step(1)
plot_dict['x'].append(_step)
plot_dict['y_loss'].append(solver.net.blobs['loss'].data[...].tolist())
# if args.debug==1:
# layer_list = ['predict_reshape', 'concat2']
plot_iter = args.plotiter
epoch = 20
idx_counter = 0
x = []
y1 = []
y2 = []
z = [] # validation
plt.plot(x, y1)
_step = 0
while _step < max_iter:
if _step % validation_iter == 0:
##do validation
pass
frame_data, density_data = tranining_dataset.get_frame_c3d(
mini_batch=batch)
solver.net.blobs['data'].data[...] = frame_data
solver.net.blobs['ground_truth'].data[...] = density_data
solver.step(1)
x.append(_step)
y1.append(solver.net.blobs['loss'].data[...].tolist())
# y2.append(solver.net.blobs['loss5'].data[...].tolist())
plt.plot(x, y1)
if _step % plot_iter == 0:
plt.xlabel('Iter')
plt.ylabel('loss')
plt.savefig(os.path.join(plot_figure_dir,
"plot" + str(_step) + ".png"))
print >> log_f, 'use test set', testset_path,
solver = caffe.AdaDeltaSolver(solver_path)
solver.net.copy_from(model_path)
test_tuple_list = pkl.load(open(testset_path, 'rb'))
print >> log_f, ' test set length', len(test_tuple_list)
# print testset_path_list, test_tuple_list;exit()
dataset.setup_video_dataset_c3d(overlap=8, training_example_props=0.9)
dataset.validation_tuple_list = test_tuple_list
# print dataset.validation_tuple_list[:10], test_tuple_list[:10]
# exit()
batch_size = 2
data_tuple = dataset.get_frame_c3d(mini_batch=batch_size,
phase='validation',
density_length='full')
tmp_cc = []
tmp_sim = []
tmp_kld = []
tmp_aucj = []
tmp_aucb = []
index = 0
while data_tuple is not None:
print index, 'of', len(test_tuple_list), '\r',
sys.stdout.flush()
frame_data, density_data = data_tuple
solver.net.blobs['data'].data[...] = frame_data
solver.net.blobs['ground_truth'].data[...] = density_data
solver.net.forward()
predictions = solver.net.blobs['predict'].data[