提交 58261d1d 编写于 作者: F Fan Yang 提交者: A. Unique TensorFlower

Internal change

PiperOrigin-RevId: 489040948
上级 a4e50484
......@@ -338,29 +338,50 @@ class DetectionTask(base_task.Task):
# Evaluator class handles loss metric for you.
logs = {self.loss: loss}
# This is for backward compatibility.
if 'detection_boxes' not in outputs:
detection_boxes = box_ops.cycxhw_to_yxyx(
outputs['box_outputs']) * tf.expand_dims(
tf.concat([
labels['image_info'][:, 1:2, 0], labels['image_info'][:, 1:2,
1],
labels['image_info'][:, 1:2, 0], labels['image_info'][:, 1:2,
1]
],
axis=1),
axis=1)
else:
detection_boxes = outputs['detection_boxes']
detection_scores = tf.math.reduce_max(
tf.nn.softmax(outputs['cls_outputs'])[:, :, 1:], axis=-1
) if 'detection_scores' not in outputs else outputs['detection_scores']
if 'detection_classes' not in outputs:
detection_classes = tf.math.argmax(
outputs['cls_outputs'][:, :, 1:], axis=-1) + 1
else:
detection_classes = outputs['detection_classes']
if 'num_detections' not in outputs:
num_detections = tf.reduce_sum(
tf.cast(
tf.math.greater(
tf.math.reduce_max(outputs['cls_outputs'], axis=-1), 0),
tf.int32),
axis=-1)
else:
num_detections = outputs['num_detections']
predictions = {
'detection_boxes':
outputs['detection_boxes'] * tf.expand_dims(
tf.concat([
labels['image_info'][:, 1:2,
0], labels['image_info'][:, 1:2, 1],
labels['image_info'][:, 1:2,
0], labels['image_info'][:, 1:2, 1]
],
axis=1),
axis=1),
'detection_scores':
outputs['detection_scores'],
'detection_classes':
outputs['detection_classes'],
# Fix this. It's not being used at the moment.
'num_detections':
outputs['num_detections'],
'source_id':
labels['id'],
'image_info':
labels['image_info']
'detection_boxes': detection_boxes,
'detection_scores': detection_scores,
'detection_classes': detection_classes,
'num_detections': num_detections,
'source_id': labels['id'],
'image_info': labels['image_info']
}
ground_truths = {
'source_id': labels['id'],
'height': labels['image_info'][:, 0:1, 0],
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册