def __loss_per_scale(self, name, conv, pred, label, bboxes, anchors,
stride):
"""
:param name: loss的名字
:param conv: conv是yolo卷积层的原始输出
shape为(batch_size, output_size, output_size, anchor_per_scale * (5 + num_class))
:param pred: conv是yolo输出的预测bbox的信息(x, y, w, h, conf, prob)，
其中(x, y, w, h)的大小是相对于input_size的，如input_size=416，(x, y, w, h) = (120, 200, 50, 70)
shape为(batch_size, output_size, output_size, anchor_per_scale, 5 + num_class)
:param label: shape为(batch_size, output_size, output_size, anchor_per_scale, 5 + num_classes)
只有best anchor对应位置的数据才为(x, y, w, h, 1, classes), (x, y, w, h)的大小是bbox纠正后的原始大小
:param bboxes: shape为(batch_size, max_bbox_per_scale, 4)，
存储的坐标为(x, y, w, h)，(x, y, w, h)的大小都是bbox纠正后的原始大小
bboxes用于计算相应detector的预测框与该detector负责预测的所有bbox的IOU
:param anchors: 相应detector的anchors
:param stride: 相应detector的stride
"""
with tf.name_scope(name):
conv_shape = tf.shape(conv)
batch_size = conv_shape[0]
output_size = conv_shape[1]
input_size = stride * output_size
conv = tf.reshape(
conv, (batch_size, output_size, output_size,
self.__anchor_per_scale, 5 + self.__num_classes))
conv_raw_dxdy = conv[:, :, :, :, 0:2]
conv_raw_dwdh = conv[:, :, :, :, 2:4]
conv_raw_conf = conv[:, :, :, :, 4:5]
conv_raw_prob = conv[:, :, :, :, 5:]
pred_xywh = pred[:, :, :, :, 0:4]
pred_conf = pred[:, :, :, :, 4:5]
label_xy = label[:, :, :, :, 0:2]
label_wh = label[:, :, :, :, 2:4]
respond_bbox = label[:, :, :, :, 4:5]
label_prob = label[:, :, :, :, 5:]
# (1)计算xywh损失
y = tf.tile(
tf.range(output_size, dtype=tf.int32)[:, tf.newaxis],
[1, output_size])
x = tf.tile(
tf.range(output_size, dtype=tf.int32)[tf.newaxis, :],
[output_size, 1])
xy_grid = tf.concat([x[:, :, tf.newaxis], y[:, :, tf.newaxis]],
axis=-1)
xy_grid = tf.tile(xy_grid[tf.newaxis, :, :, tf.newaxis, :],
[batch_size, 1, 1, self.__anchor_per_scale, 1])
xy_grid = tf.cast(xy_grid, tf.float32)
label_txty = 1.0 * label_xy / stride - xy_grid
label_raw_twth = tf.log((1.0 * label_wh / stride) / anchors)
label_raw_twth = tf.where(tf.is_inf(label_raw_twth),
tf.zeros_like(label_raw_twth),
label_raw_twth)
input_size = tf.cast(input_size, tf.float32)
bbox_loss_scale = 2.0 - 1.0 * label_wh[:, :, :, :, 0:
1] * label_wh[:, :, :, :,
1:2] / (
input_size
**2)
xy_loss = respond_bbox * bbox_loss_scale * \
tf.nn.sigmoid_cross_entropy_with_logits(labels=label_txty, logits=conv_raw_dxdy)
wh_loss = 0.5 * respond_bbox * bbox_loss_scale * tf.square(
label_raw_twth - conv_raw_dwdh)
# (2)计算confidence损失
iou = utils.iou_calc4(
pred_xywh[:, :, :, :, np.newaxis, :],
bboxes[:, np.newaxis, np.newaxis, np.newaxis, :, :])
max_iou = tf.reduce_max(iou, axis=-1)
max_iou = max_iou[:, :, :, :, np.newaxis]
respond_bgd = (1.0 - respond_bbox) * tf.cast(
max_iou < self.__iou_loss_thresh, tf.float32)
conf_focal = self.__focal(respond_bbox, pred_conf)
conf_loss = conf_focal * (
respond_bbox * tf.nn.sigmoid_cross_entropy_with_logits(
labels=respond_bbox, logits=conv_raw_conf) +
respond_bgd * tf.nn.sigmoid_cross_entropy_with_logits(
labels=respond_bbox, logits=conv_raw_conf))
# (3)计算classes损失
prob_loss = respond_bbox * tf.nn.sigmoid_cross_entropy_with_logits(
labels=label_prob, logits=conv_raw_prob)
loss = tf.concat([xy_loss, wh_loss, conf_loss, prob_loss], axis=-1)
loss = tf.reduce_mean(tf.reduce_sum(loss, axis=[1, 2, 3, 4]))
return loss
def __loss_per_scale(self, name, conv, pred, label, bboxes, anchors,
stride):
"""
:param name: loss的名字
:param conv: conv是yolo卷积层的原始输出
shape为(batch_size, output_size, output_size, anchor_per_scale * (5 + num_class))
:param pred: conv是yolo输出的预测bbox的信息(x, y, w, h, conf, prob)，
其中(x, y, w, h)的大小是相对于input_size的，如input_size=416，(x, y, w, h) = (120, 200, 50, 70)
shape为(batch_size, output_size, output_size, anchor_per_scale, 5 + num_class)
:param label: shape为(batch_size, output_size, output_size, anchor_per_scale, 5 + num_classes)
只有best anchor对应位置的数据才为(x, y, w, h, 1, classes), (x, y, w, h)的大小是bbox纠正后的原始大小
:param bboxes: shape为(batch_size, max_bbox_per_scale, 4)，
存储的坐标为(x, y, w, h)，(x, y, w, h)的大小都是bbox纠正后的原始大小
bboxes用于计算相应detector的预测框与该detector负责预测的所有bbox的IOU
:param anchors: 相应detector的anchors
:param stride: 相应detector的stride
"""
with tf.name_scope(name):
conv_shape = tf.shape(conv)
batch_size = conv_shape[0]
output_size = conv_shape[1]
input_size = stride * output_size
conv = tf.reshape(
conv, (batch_size, output_size, output_size,
self.__anchor_per_scale, 5 + self.__num_classes))
conv_raw_conf = conv[:, :, :, :, 4:5]
conv_raw_prob = conv[:, :, :, :, 5:]
pred_xywh = pred[:, :, :, :, 0:4]
pred_conf = pred[:, :, :, :, 4:5]
label_xywh = label[:, :, :, :, 0:4]
respond_bbox = label[:, :, :, :, 4:5]
label_prob = label[:, :, :, :, 5:]
GIOU = utils.GIOU(pred_xywh, label_xywh)
GIOU = GIOU[..., np.newaxis]
input_size = tf.cast(input_size, tf.float32)
bbox_loss_scale = 2.0 - 1.0 * label_xywh[:, :, :, :, 2:
3] * label_xywh[:, :, :, :,
3:4] / (
input_size
**2)
GIOU_loss = respond_bbox * bbox_loss_scale * (1.0 - GIOU)
# (2)计算confidence损失
iou = utils.iou_calc4(
pred_xywh[:, :, :, :, np.newaxis, :],
bboxes[:, np.newaxis, np.newaxis, np.newaxis, :, :])
max_iou = tf.reduce_max(iou, axis=-1)
max_iou = max_iou[:, :, :, :, np.newaxis]
respond_bgd = (1.0 - respond_bbox) * tf.cast(
max_iou < self.__iou_loss_thresh, tf.float32)
conf_focal = self.__focal(respond_bbox, pred_conf)
conf_loss = conf_focal * (
respond_bbox * tf.nn.sigmoid_cross_entropy_with_logits(
labels=respond_bbox, logits=conv_raw_conf) +
respond_bgd * tf.nn.sigmoid_cross_entropy_with_logits(
labels=respond_bbox, logits=conv_raw_conf))
# (3)计算classes损失
prob_loss = respond_bbox * tf.nn.sigmoid_cross_entropy_with_logits(
labels=label_prob, logits=conv_raw_prob)
loss = tf.concat([GIOU_loss, conf_loss, prob_loss], axis=-1)
loss = tf.reduce_mean(tf.reduce_sum(loss, axis=[1, 2, 3, 4]))
return loss