def train_mscoco():
# ===========================模型的配置和加载======================================
# config for data argument
cfg = config.Config()
cfg.use_horizontal_flips = True
cfg.use_vertical_flips = True
cfg.rot_90 = True
cfg.num_rois = 32
#resnet前四卷积部分的权值
cfg.base_net_weights = nn.get_weight_path()
#保存模型的权重值
cfg.model_path = './model/mscoco_frcnn.hdf5'
#all_images, class_mapping = get_data()
#加载训练的图片
train_imgs, class_mapping = get_data('train')
cfg.class_mapping = class_mapping
print('Num classes (including bg) = {}'.format(len(class_mapping)))
#保存所有的配置文件
with open(cfg.config_save_file, 'wb') as config_f:
pickle.dump(cfg, config_f)
print(
'Config has been written to {}, and can be loaded when testing to ensure correct results'
.format(cfg.config_save_file))
#图片随机洗牌
random.shuffle(train_imgs)
print('Num train samples {}'.format(len(train_imgs)))
data_gen_train = data_generators.get_anchor_gt(train_imgs,
class_mapping,
cfg,
nn.get_img_output_length,
K.image_dim_ordering(),
mode='train')
# ==============================================================================
# ===============================模型的定义======================================
#keras内核为tensorflow
input_shape_img = (None, None, 3)
img_input = Input(shape=input_shape_img)
roi_input = Input(shape=(None, 4))
# define the base resnet50 network
shared_layers = nn.nn_base(img_input, trainable=False)
# define the RPN, built on the base layers
num_anchors = len(cfg.anchor_box_scales) * len(cfg.anchor_box_ratios)
rpn = nn.rpn(shared_layers, num_anchors)
classifier = nn.classifier(shared_layers,
roi_input,
cfg.num_rois,
nb_classes=len(class_mapping),
trainable=True)
#model(input=,output=)
model_rpn = Model(img_input, rpn[:2])
model_classifier = Model([img_input, roi_input], classifier)
# this is a model that holds both the RPN and the classifier, used to load/save weights for the models
model_all = Model([img_input, roi_input], rpn[:2] + classifier)
# ==============================================================================
# ===========================基本模型加载ImageNet权值=============================
try:
print('loading base model weights from {}'.format(
cfg.base_net_weights))
model_rpn.load_weights(cfg.base_net_weights, by_name=True)
model_classifier.load_weights(cfg.base_net_weights, by_name=True)
except Exception as e:
print('基本模型加载ImageNet权值: ', e)
print('Could not load pretrained model weights on ImageNet.')
# ==============================================================================
# ===============================模型优化========================================
#在调用model.compile()之前初始化一个优化器对象，然后传入该函数
optimizer = Adam(lr=1e-5)
optimizer_classifier = Adam(lr=1e-5)
model_rpn.compile(optimizer=optimizer,
loss=[
losses_fn.rpn_loss_cls(num_anchors),
losses_fn.rpn_loss_regr(num_anchors)
])
model_classifier.compile(
optimizer=optimizer_classifier,
loss=[
losses_fn.class_loss_cls,
losses_fn.class_loss_regr(len(class_mapping) - 1)
],
metrics={'dense_class_{}'.format(len(class_mapping)): 'accuracy'})
model_all.compile(optimizer='sgd', loss='mae')
# ==============================================================================
# ================================训练、输出设置==================================
epoch_length = len(train_imgs)
num_epochs = int(cfg.num_epochs)
iter_num = 0
losses = np.zeros((epoch_length, 5))
rpn_accuracy_rpn_monitor = []
rpn_accuracy_for_epoch = []
start_time = time.time()
best_loss = np.Inf
logger = Logger(os.path.join('.', 'log.txt'))
# ==============================================================================
print('Starting training')
for epoch_num in range(num_epochs):
progbar = generic_utils.Progbar(epoch_length)
logger.write('Epoch {}/{}'.format(epoch_num + 1, num_epochs))
while True:
try:
if len(rpn_accuracy_rpn_monitor
) == epoch_length and cfg.verbose:
mean_overlapping_bboxes = float(
sum(rpn_accuracy_rpn_monitor)) / len(
rpn_accuracy_rpn_monitor)
rpn_accuracy_rpn_monitor = []
print(
'Average number of overlapping bounding boxes from RPN = {} for {} previous iterations'
.format(mean_overlapping_bboxes, epoch_length))
if mean_overlapping_bboxes == 0:
print(
'RPN is not producing bounding boxes that overlap'
' the ground truth boxes. Check RPN settings or keep training.'
)
#图片，标准的cls、rgr，盒子数据
X, Y, img_data = next(data_gen_train)
#训练rpn
loss_rpn = model_rpn.train_on_batch(X, Y)
#边训练rpn得到的区域送入roi
#x_class, x_regr, base_layers
P_rpn = model_rpn.predict_on_batch(X)
result = roi_helpers.rpn_to_roi(P_rpn[0],
P_rpn[1],
cfg,
K.image_dim_ordering(),
use_regr=True,
overlap_thresh=0.7,
max_boxes=300)
# note: calc_iou converts from (x1,y1,x2,y2) to (x,y,w,h) format
#区域、cls、rgr、iou
X2, Y1, Y2, IouS = roi_helpers.calc_iou(
result, img_data, cfg, class_mapping)
if X2 is None:
rpn_accuracy_rpn_monitor.append(0)
rpn_accuracy_for_epoch.append(0)
continue
neg_samples = np.where(Y1[0, :, -1] == 1)
pos_samples = np.where(Y1[0, :, -1] == 0)
if len(neg_samples) > 0:
neg_samples = neg_samples[0]
else:
neg_samples = []
if len(pos_samples) > 0:
pos_samples = pos_samples[0]
else:
pos_samples = []
rpn_accuracy_rpn_monitor.append(len(pos_samples))
rpn_accuracy_for_epoch.append((len(pos_samples)))
if cfg.num_rois > 1:
if len(pos_samples) < cfg.num_rois // 2:
selected_pos_samples = pos_samples.tolist()
else:
selected_pos_samples = np.random.choice(
pos_samples, cfg.num_rois // 2,
replace=False).tolist()
try:
selected_neg_samples = np.random.choice(
neg_samples,
cfg.num_rois - len(selected_pos_samples),
replace=False).tolist()
except:
selected_neg_samples = np.random.choice(
neg_samples,
cfg.num_rois - len(selected_pos_samples),
replace=True).tolist()
sel_samples = selected_pos_samples + selected_neg_samples
else:
# in the extreme case where num_rois = 1, we pick a random pos or neg sample
selected_pos_samples = pos_samples.tolist()
selected_neg_samples = neg_samples.tolist()
if np.random.randint(0, 2):
sel_samples = random.choice(neg_samples)
else:
sel_samples = random.choice(pos_samples)
#训练classifier
loss_class = model_classifier.train_on_batch(
[X, X2[:, sel_samples, :]],
[Y1[:, sel_samples, :], Y2[:, sel_samples, :]])
losses[iter_num, 0] = loss_rpn[1]
losses[iter_num, 1] = loss_rpn[2]
losses[iter_num, 2] = loss_class[1]
losses[iter_num, 3] = loss_class[2]
losses[iter_num, 4] = loss_class[3]
iter_num += 1
progbar.update(
iter_num,
[('rpn_cls', np.mean(losses[:iter_num, 0])),
('rpn_regr', np.mean(losses[:iter_num, 1])),
('detector_cls', np.mean(losses[:iter_num, 2])),
('detector_regr', np.mean(losses[:iter_num, 3]))])
if iter_num == epoch_length:
loss_rpn_cls = np.mean(losses[:, 0])
loss_rpn_regr = np.mean(losses[:, 1])
loss_class_cls = np.mean(losses[:, 2])
loss_class_regr = np.mean(losses[:, 3])
class_acc = np.mean(losses[:, 4])
mean_overlapping_bboxes = float(sum(
rpn_accuracy_for_epoch)) / len(rpn_accuracy_for_epoch)
rpn_accuracy_for_epoch = []
if cfg.verbose:
logger.write(
'Mean number of bounding boxes from RPN overlapping ground truth boxes: {}'
.format(mean_overlapping_bboxes))
logger.write(
'Classifier accuracy for bounding boxes from RPN: {}'
.format(class_acc))
logger.write(
'Loss RPN classifier: {}'.format(loss_rpn_cls))
logger.write(
'Loss RPN regression: {}'.format(loss_rpn_regr))
logger.write('Loss Detector classifier: {}'.format(
loss_class_cls))
logger.write('Loss Detector regression: {}'.format(
loss_class_regr))
logger.write('Elapsed time: {}'.format(time.time() -
start_time))
curr_loss = loss_rpn_cls + loss_rpn_regr + loss_class_cls + loss_class_regr
iter_num = 0
start_time = time.time()
if curr_loss < best_loss:
if cfg.verbose:
logger.write(
'Total loss decreased from {} to {}, saving weights'
.format(best_loss, curr_loss))
best_loss = curr_loss
model_all.save_weights(cfg.model_path)
break
except Exception as e:
print('Exception: {}'.format(e))
# save model
model_all.save_weights(cfg.model_path)
continue
print('Training complete, exiting.')
# 생각, 계산 최소화. [(유효한 길이 합) * VOCAB_SIZE] 주의할 점은 테스트뽑을때 배치별로 붙으므로 한줄을 읽으려면 stride를 BATCH_SIZE로 줘야한다.
target = nn.utils.rnn.pack_padded_sequence(
target, pred_length,
batch_first=True) # 마찬가지로 패딩을 모두 없애고 한줄로 핀다. [유효한 길이 합]
loss = criterion(preds.data, target.data)
loss = loss + args.ATTENTION_COEF * (
(1.0 - coefs.sum(dim=1))**2).mean() # regularization
train_loss = train_loss + loss
loss.backward()
decoder_optimizer.step()
encoder_optimizer.step() # GAN과 마찬가지로 역순
train_losses.append(train_loss)
if idx % 100 == 0:
logger.write("batch : [{}/{}]\n".format(
idx, len(train_loader)))
logger.write("train_loss : {}\n".format(
train_loss / (len(train_loader) * args.NUM_CAPTIONS)))
logger.write("-" * 10 + "validation" + "-" * 10 + "\n")
encoder.eval()
decoder.eval()
with torch.no_grad():
for idx, (img, caption_5, caption_lengths_5,
_) in enumerate(validation_loader):
origin_img = img
for i in range(args.NUM_CAPTIONS):
img = origin_img.to(device)
caption = caption_5[:, i, :].to(device)
caption_lengths = caption_lengths_5[:, :, i].to(device)