Adds histograms of the tensor's nonzero values for each scale (feature map). Arguments: tensor: a float tensor with shape [batch_

Computes area of boxes. Arguments: boxes: a float tensor with shape [N, 4]. Returns: a float tensor with shape [N] representi

Convert bounding boxes of the format [ymin, xmin, ymax, xmax] to the format [cy, cx, h, w]. Arguments: boxes: a list of float tensors

Adds histograms of the biggest 20 percent of tensor's values for each scale (feature map). Arguments: tensor: a float ten

Decodes a batch of box encodings with respect to the anchors. Arguments: box_encodings: a float tensor with shape [batch_size, num_anchor

Compute pairwise intersection areas between boxes. Arguments: boxes1: a float tensor with shape [N, 4]. boxes2: a float tensor wi

Convert bounding boxes of the format [cy, cx, h, w] to the format [ymin, xmin, ymax, xmax]. Arguments: boxes: a list of float tensors

Adds box predictors to each feature map, reshapes, and returns concatenated results. Arguments: feature_maps: a list of float ten

Adds summaries for the number of matches on each scale.

Change coordinate frame of the boxes to be relative to window's frame. Arguments: boxes: a float tensor with shape [N, 4]. window

Returns regression targets for each anchor. Arguments: anchors: a float tensor with shape [num_anchors, 4]. groundtruth_boxes: a

Arguments: groundtruth: a dict with the following keys 'boxes': a float tensor with shape [batch_size, N, 4].

Computes pairwise intersection-over-area between box collections. intersection-over-area (IOA) between two boxes box1 and box2 is defined as t

Matching algorithm: 1) for each groundtruth box choose the anchor with largest iou, 2) remove this set of anchors from the set of all anchors,

Prunes bounding boxes that fall completely outside of the given window. This function does not clip partially overflowing boxes. Arguments:

Prunes the boxes in boxes1 that overlap less than thresh with boxes2. For each box in boxes1, we want its IOA to be more than min_overlap with

Performs random crop. Given the input image and its bounding boxes, this op randomly crops a subimage. Given a user-provided set of input constra

Subsample a collection of selected indices to a desired neg:pos ratio. Arguments: indices: an int or long tensor with shape [M],

Arguments: images: a numpy string array with shape []. boxes: a numpy float array with shape [N, 4]. scor

Add L2 regularization to all (or some) trainable kernel weights.

Applies hard mining to anchorwise losses. Arguments: location_losses: a float tensor with shape [batch_size, num_anchors]. cls_lo

Arguments: boxes: a float tensor with shape [batch_size, N, 4]. scores: a float tensor with shape [batch_size, N]. score_

Arguments: predictions: a float tensor with shape [batch_size, num_anchors, num_classes + 1], representing the predicted logi

Arguments: precision, recall, confidences: lists of floats of the same length. Returns: 1. a float number, best confidence t

Decode relative codes to boxes. Arguments: codes: a float tensor with shape [N, 4], anchor-encoded boxes of the format [ty, t

Convert dict to tf.Example proto. Notice that this function normalizes the bounding box coordinates provided by the raw data. Arguments:

Encode boxes with respect to anchors. Arguments: boxes: a float tensor with shape [N, 4]. anchors: a float tensor with shape [N,

Arguments: groundtruth_by_img: a dict of lists with boxes, image -> list of groundtruth boxes on the image. all_detec

Generate densified anchor boxes at (0, 0) grid position.

Returns: features: a dict with the following keys 'images': a float tensor with shape [batch_size, image_height,

Arguments: image_name: a string tensor with shape [1]. groundtruth: a dict with the following keys 'b

Postprocess outputs of the network. Returns: boxes: a float tensor with shape [batch_size, N, 4]. scores: a float ten

Arguments: anchors: a float tensor with shape [num_anchors, 4]. groundtruth_boxes: a float tensor with shape [N, 4]. thre

Computes pairwise intersection-over-union between two box collections. Arguments: boxes1: a float tensor with shape [N, 4]. boxes

A usual L1 smooth loss. Arguments: predictions: a float tensor with shape [batch_size, num_anchors, 4], representing the (enc

Compute scalar loss tensors with respect to provided groundtruth. Arguments: groundtruth: a dict with the following keys

Arguments: detection: a box. groundtruth_boxes: a list of boxes. Returns: best_i: an integer, index of the best groun

Transform images before feeding them to the network.

Randomly jitter a box. Arguments: box: a float tensor with shape [4]. ratio: a float number. Returns:

Randomly jitter bounding boxes. Arguments: boxes: a float tensor with shape [N, 4]. ratio: a float number. The ratio

This function scales each pixel independently of the other ones. Arguments: image: a float tensor with shape [height, width, 3],

Arguments: image_size: a tuple of integers (width, height). grid_height: an integer, size of the grid in the y direction.

Detect faces. Arguments: image: a numpy uint8 array with shape [height, width, 3], that represents a RGB image.

Arguments: images: a float tensor with shape [batch_size, height, width, 3], a batch of RGB images with pixel val

Arguments: image_features: a list of float tensors where the ith tensor has shape [batch, height_i, width_i, chan

Arguments: image: a string, identifier of a image. box: a numpy float array with shape [4]. score: a floa

Arguments: model_path: a string, path to a pb file. gpu_memory_fraction: a float number. visible_device_l

Arguments: images: a float tensor with shape [batch_size, height, width, 3], a batch of RGB images with pixel val

Note: when evaluating set batch_size to 1. Arguments: filenames: a list of strings, paths to tfrecords files.

What this function does: 1. Parses one record from a tfrecords file and decodes it. 2. (optionally) Augments it. Returns:

This is a function for creating a computational tensorflow graph. The function is in format required by tf.estimator.