def pycaffe_predict(im):
prototxt_file = 'prototxt/32/predict.prototxt'
weights_file = 'predict.caffemodel'
if caffe.is_initialized() < 3:
caffe.init(prototxt_file, weights_file)
caffe.set_device(0)
im = im.astype('float32')
im[:, :, 0] = im[:, :, 0] - 104.008
im[:, :, 1] = im[:, :, 1] - 116.669
im[:, :, 2] = im[:, :, 2] - 122.675
im = np.expand_dims(im, 3)
im = np.transpose(im, (3, 2, 0, 1)).copy()
input_data = [im]
t1 = time.time()
score = caffe.forward(2, input_data)
t2 = time.time()
print t2 - t1
raw_score = score[0]
raw_score = raw_score[0, :, :, :]
raw_score = np.transpose(raw_score, (1, 2, 0)).copy()
raw_score[:, :, 0] = raw_score[:, :, 0] + 104.008
raw_score[:, :, 1] = raw_score[:, :, 1] + 116.669
raw_score[:, :, 2] = raw_score[:, :, 2] + 122.675
raw_score = np.clip(raw_score, 0, 255)
return raw_score.astype('uint8')
def pycaffe_param(hidden_feat):
prototxt_file = 'prototxt/32/param.prototxt'
weights_file = 'model/train_32.caffemodel'
if caffe.is_initialized() < 2:
caffe.init(prototxt_file, weights_file)
caffe.set_device(0)
hidden_feat = hidden_feat.reshape((1, hidden_feat.size, 1, 1))
input_data = [hidden_feat]
param = caffe.forward(1, input_data)
caffe.save_model(param, 'layer_name.txt', 'base.caffemodel',
'predict.caffemodel')
def pycaffe_hidden(im_label):
prototxt_file = 'prototxt/32/hidden.prototxt'
weights_file = 'model/train_32.caffemodel'
if caffe.is_initialized() < 1:
caffe.init(prototxt_file, weights_file)
caffe.set_device(0)
im_label = im_label.astype('float32')
im_label[:, :, 0] = im_label[:, :, 0] - 104.008
im_label[:, :, 1] = im_label[:, :, 1] - 116.669
im_label[:, :, 2] = im_label[:, :, 2] - 122.675
im_label_ = np.expand_dims(im_label, 3)
im_label = np.transpose(im_label_, (3, 2, 0, 1)).copy()
input_data = [im_label]
score = caffe.forward(0, input_data)
hidden_feat = score[0].squeeze()
return hidden_feat