def testGetOneMAPWithMatchingGroundtruthAndDetectionsSkipCrowd(self):
"""Tests computing mAP with is_crowd GT boxes skipped."""
category_list = [{
'id': 0,
'name': 'person'
}, {
'id': 1,
'name': 'cat'
}, {
'id': 2,
'name': 'dog'
}]
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(category_list)
coco_evaluator.add_single_ground_truth_image_info(
image_id='image1',
groundtruth_dict={
standard_fields.InputDataFields.groundtruth_boxes:
np.array([[100., 100., 200., 200.], [99., 99., 200., 200.]]),
standard_fields.InputDataFields.groundtruth_classes:
np.array([1, 2]),
standard_fields.InputDataFields.groundtruth_is_crowd:
np.array([0, 1])
})
coco_evaluator.add_single_detected_image_info(
image_id='image1',
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[100., 100., 200., 200.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
metrics = coco_evaluator.evaluate()
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/mAP'], 1.0)
def testRejectionOnDuplicateDetections(self):
"""Tests that detections cannot be added more than once for an image."""
categories = [{
'id': 1,
'name': 'cat'
}, {
'id': 2,
'name': 'dog'
}, {
'id': 3,
'name': 'elephant'
}]
# Add groundtruth
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(categories)
coco_evaluator.add_single_ground_truth_image_info(
image_id='image1',
groundtruth_dict={
standard_fields.InputDataFields.groundtruth_boxes:
np.array([[99., 100., 200., 200.]]),
standard_fields.InputDataFields.groundtruth_classes:
np.array([1])
})
coco_evaluator.add_single_detected_image_info(
image_id='image1',
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[100., 100., 200., 200.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
detections_lists_len = len(coco_evaluator._detection_boxes_list)
coco_evaluator.add_single_detected_image_info(
image_id='image1', # Note that this image id was previously added.
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[100., 100., 200., 200.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
self.assertEqual(detections_lists_len,
len(coco_evaluator._detection_boxes_list))
def testRejectionOnDuplicateGroundtruth(self):
"""Tests that groundtruth cannot be added more than once for an image."""
categories = [{
'id': 1,
'name': 'cat'
}, {
'id': 2,
'name': 'dog'
}, {
'id': 3,
'name': 'elephant'
}]
# Add groundtruth
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(categories)
image_key1 = 'img1'
groundtruth_boxes1 = np.array(
[[0, 0, 1, 1], [0, 0, 2, 2], [0, 0, 3, 3]], dtype=float)
groundtruth_class_labels1 = np.array([1, 3, 1], dtype=int)
coco_evaluator.add_single_ground_truth_image_info(
image_key1, {
standard_fields.InputDataFields.groundtruth_boxes:
groundtruth_boxes1,
standard_fields.InputDataFields.groundtruth_classes:
groundtruth_class_labels1
})
groundtruth_lists_len = len(coco_evaluator._groundtruth_list)
# Add groundtruth with the same image id.
coco_evaluator.add_single_ground_truth_image_info(
image_key1, {
standard_fields.InputDataFields.groundtruth_boxes:
groundtruth_boxes1,
standard_fields.InputDataFields.groundtruth_classes:
groundtruth_class_labels1
})
self.assertEqual(groundtruth_lists_len,
len(coco_evaluator._groundtruth_list))
def testExceptionRaisedWithMissingGroundtruth(self):
"""Tests that exception is raised for detection with missing groundtruth."""
categories = [{
'id': 1,
'name': 'cat'
}, {
'id': 2,
'name': 'dog'
}, {
'id': 3,
'name': 'elephant'
}]
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(categories)
with self.assertRaises(ValueError):
coco_evaluator.add_single_detected_image_info(
image_id='image1',
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[100., 100., 200., 200.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
def model_fn(features, labels, mode, params=None):
"""Constructs the object detection model.
Args:
features: Dictionary of feature tensors, returned from `input_fn`.
labels: Dictionary of groundtruth tensors if mode is TRAIN or EVAL,
otherwise None.
mode: Mode key from tf.estimator.ModeKeys.
params: Parameter dictionary passed from the estimator.
Returns:
An `EstimatorSpec` that encapsulates the model and its serving
configurations.
"""
params = params or {}
total_loss, train_op, detections, export_outputs = None, None, None, None
is_training = mode == tf.estimator.ModeKeys.TRAIN
detection_model = detection_model_fn(is_training=is_training,
add_summaries=(not use_tpu))
scaffold_fn = None
if mode == tf.estimator.ModeKeys.TRAIN:
labels = unstack_batch(
labels,
unpad_groundtruth_tensors=train_config.unpad_groundtruth_tensors)
elif mode == tf.estimator.ModeKeys.EVAL:
labels = unstack_batch(labels, unpad_groundtruth_tensors=False)
if mode in (tf.estimator.ModeKeys.TRAIN, tf.estimator.ModeKeys.EVAL):
gt_boxes_list = labels[fields.InputDataFields.groundtruth_boxes]
gt_classes_list = labels[fields.InputDataFields.groundtruth_classes]
gt_masks_list = None
if fields.InputDataFields.groundtruth_instance_masks in labels:
gt_masks_list = labels[
fields.InputDataFields.groundtruth_instance_masks]
gt_keypoints_list = None
if fields.InputDataFields.groundtruth_keypoints in labels:
gt_keypoints_list = labels[fields.InputDataFields.groundtruth_keypoints]
detection_model.provide_groundtruth(
groundtruth_boxes_list=gt_boxes_list,
groundtruth_classes_list=gt_classes_list,
groundtruth_masks_list=gt_masks_list,
groundtruth_keypoints_list=gt_keypoints_list)
preprocessed_images = features[fields.InputDataFields.image]
prediction_dict = detection_model.predict(
preprocessed_images, features[fields.InputDataFields.true_image_shape])
detections = detection_model.postprocess(
prediction_dict, features[fields.InputDataFields.true_image_shape])
if mode == tf.estimator.ModeKeys.TRAIN:
if train_config.fine_tune_checkpoint and hparams.load_pretrained:
asg_map = detection_model.restore_map(
from_detection_checkpoint=train_config.from_detection_checkpoint,
load_all_detection_checkpoint_vars=(
train_config.load_all_detection_checkpoint_vars))
available_var_map = (
variables_helper.get_variables_available_in_checkpoint(
asg_map, train_config.fine_tune_checkpoint,
include_global_step=False))
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(train_config.fine_tune_checkpoint,
available_var_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(train_config.fine_tune_checkpoint,
available_var_map)
if mode in (tf.estimator.ModeKeys.TRAIN, tf.estimator.ModeKeys.EVAL):
losses_dict = detection_model.loss(
prediction_dict, features[fields.InputDataFields.true_image_shape])
losses = [loss_tensor for loss_tensor in losses_dict.itervalues()]
total_loss = tf.add_n(losses, name='total_loss')
if mode == tf.estimator.ModeKeys.TRAIN:
global_step = tf.train.get_or_create_global_step()
training_optimizer, optimizer_summary_vars = optimizer_builder.build(
train_config.optimizer)
if use_tpu:
training_optimizer = tpu_optimizer.CrossShardOptimizer(
training_optimizer)
# Optionally freeze some layers by setting their gradients to be zero.
trainable_variables = None
if train_config.freeze_variables:
trainable_variables = tf.contrib.framework.filter_variables(
tf.trainable_variables(),
exclude_patterns=train_config.freeze_variables)
clip_gradients_value = None
if train_config.gradient_clipping_by_norm > 0:
clip_gradients_value = train_config.gradient_clipping_by_norm
if not use_tpu:
for var in optimizer_summary_vars:
tf.summary.scalar(var.op.name, var)
summaries = [] if use_tpu else None
train_op = tf.contrib.layers.optimize_loss(
loss=total_loss,
global_step=global_step,
learning_rate=None,
clip_gradients=clip_gradients_value,
optimizer=training_optimizer,
variables=trainable_variables,
summaries=summaries,
name='') # Preventing scope prefix on all variables.
if mode == tf.estimator.ModeKeys.PREDICT:
export_outputs = {
tf.saved_model.signature_constants.PREDICT_METHOD_NAME:
tf.estimator.export.PredictOutput(detections)
}
eval_metric_ops = None
if mode == tf.estimator.ModeKeys.EVAL:
# Detection summaries during eval.
class_agnostic = (fields.DetectionResultFields.detection_classes
not in detections)
groundtruth = _get_groundtruth_data(detection_model, class_agnostic)
eval_dict = eval_util.result_dict_for_single_example(
tf.expand_dims(features[fields.InputDataFields.original_image][0], 0),
features[inputs.HASH_KEY][0],
detections,
groundtruth,
class_agnostic=class_agnostic,
scale_to_absolute=False)
if class_agnostic:
category_index = label_map_util.create_class_agnostic_category_index()
else:
category_index = label_map_util.create_category_index_from_labelmap(
eval_input_config.label_map_path)
detection_and_groundtruth = vis_utils.draw_side_by_side_evaluation_image(
eval_dict, category_index, max_boxes_to_draw=20, min_score_thresh=0.2)
if not use_tpu:
tf.summary.image('Detections_Left_Groundtruth_Right',
detection_and_groundtruth)
# Eval metrics on a single image.
detection_fields = fields.DetectionResultFields()
input_data_fields = fields.InputDataFields()
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(
category_index.values())
eval_metric_ops = coco_evaluator.get_estimator_eval_metric_ops(
image_id=eval_dict[input_data_fields.key],
groundtruth_boxes=eval_dict[input_data_fields.groundtruth_boxes],
groundtruth_classes=eval_dict[input_data_fields.groundtruth_classes],
detection_boxes=eval_dict[detection_fields.detection_boxes],
detection_scores=eval_dict[detection_fields.detection_scores],
detection_classes=eval_dict[detection_fields.detection_classes])
if use_tpu:
return tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
scaffold_fn=scaffold_fn,
predictions=detections,
loss=total_loss,
train_op=train_op,
eval_metrics=eval_metric_ops,
export_outputs=export_outputs)
else:
return tf.estimator.EstimatorSpec(
mode=mode,
predictions=detections,
loss=total_loss,
train_op=train_op,
eval_metric_ops=eval_metric_ops,
export_outputs=export_outputs)
def testGetOneMAPWithMatchingGroundtruthAndDetections(self):
category_list = [{
'id': 0,
'name': 'person'
}, {
'id': 1,
'name': 'cat'
}, {
'id': 2,
'name': 'dog'
}]
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(category_list)
image_id = tf.placeholder(tf.string, shape=())
groundtruth_boxes = tf.placeholder(tf.float32, shape=(None, 4))
groundtruth_classes = tf.placeholder(tf.float32, shape=(None))
detection_boxes = tf.placeholder(tf.float32, shape=(None, 4))
detection_scores = tf.placeholder(tf.float32, shape=(None))
detection_classes = tf.placeholder(tf.float32, shape=(None))
eval_metric_ops = coco_evaluator.get_estimator_eval_metric_ops(
image_id, groundtruth_boxes, groundtruth_classes, detection_boxes,
detection_scores, detection_classes)
_, update_op = eval_metric_ops['DetectionBoxes_Precision/mAP']
with self.test_session() as sess:
sess.run(update_op,
feed_dict={
image_id: 'image1',
groundtruth_boxes: np.array([[100., 100., 200.,
200.]]),
groundtruth_classes: np.array([1]),
detection_boxes: np.array([[100., 100., 200., 200.]]),
detection_scores: np.array([.8]),
detection_classes: np.array([1])
})
sess.run(update_op,
feed_dict={
image_id: 'image2',
groundtruth_boxes: np.array([[50., 50., 100., 100.]]),
groundtruth_classes: np.array([3]),
detection_boxes: np.array([[50., 50., 100., 100.]]),
detection_scores: np.array([.7]),
detection_classes: np.array([3])
})
sess.run(update_op,
feed_dict={
image_id: 'image3',
groundtruth_boxes: np.array([[25., 25., 50., 50.]]),
groundtruth_classes: np.array([2]),
detection_boxes: np.array([[25., 25., 50., 50.]]),
detection_scores: np.array([.9]),
detection_classes: np.array([2])
})
metrics = {}
for key, (value_op, _) in eval_metric_ops.iteritems():
metrics[key] = value_op
metrics = sess.run(metrics)
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/mAP'], 1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/[email protected]'],
1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/[email protected]'],
1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/mAP (large)'],
1.0)
self.assertAlmostEqual(
metrics['DetectionBoxes_Precision/mAP (medium)'], -1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/mAP (small)'],
1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Recall/AR@1'], 1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Recall/AR@10'], 1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Recall/AR@100'], 1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Recall/AR@100 (large)'],
1.0)
self.assertAlmostEqual(
metrics['DetectionBoxes_Recall/AR@100 (medium)'], -1.0)
self.assertAlmostEqual(metrics['DetectionBoxes_Recall/AR@100 (small)'],
1.0)
self.assertFalse(coco_evaluator._groundtruth_list)
self.assertFalse(coco_evaluator._detection_boxes_list)
self.assertFalse(coco_evaluator._image_ids)
def testGetOneMAPWithMatchingGroundtruthAndDetections(self):
"""Tests that mAP is calculated correctly on GT and Detections."""
category_list = [{
'id': 0,
'name': 'person'
}, {
'id': 1,
'name': 'cat'
}, {
'id': 2,
'name': 'dog'
}]
coco_evaluator = coco_evaluation.CocoDetectionEvaluator(category_list)
coco_evaluator.add_single_ground_truth_image_info(
image_id='image1',
groundtruth_dict={
standard_fields.InputDataFields.groundtruth_boxes:
np.array([[100., 100., 200., 200.]]),
standard_fields.InputDataFields.groundtruth_classes:
np.array([1])
})
coco_evaluator.add_single_detected_image_info(
image_id='image1',
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[100., 100., 200., 200.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
coco_evaluator.add_single_ground_truth_image_info(
image_id='image2',
groundtruth_dict={
standard_fields.InputDataFields.groundtruth_boxes:
np.array([[50., 50., 100., 100.]]),
standard_fields.InputDataFields.groundtruth_classes:
np.array([1])
})
coco_evaluator.add_single_detected_image_info(
image_id='image2',
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[50., 50., 100., 100.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
coco_evaluator.add_single_ground_truth_image_info(
image_id='image3',
groundtruth_dict={
standard_fields.InputDataFields.groundtruth_boxes:
np.array([[25., 25., 50., 50.]]),
standard_fields.InputDataFields.groundtruth_classes:
np.array([1])
})
coco_evaluator.add_single_detected_image_info(
image_id='image3',
detections_dict={
standard_fields.DetectionResultFields.detection_boxes:
np.array([[25., 25., 50., 50.]]),
standard_fields.DetectionResultFields.detection_scores:
np.array([.8]),
standard_fields.DetectionResultFields.detection_classes:
np.array([1])
})
metrics = coco_evaluator.evaluate()
self.assertAlmostEqual(metrics['DetectionBoxes_Precision/mAP'], 1.0)