def __call__(self, box_outputs, class_outputs, anchor_boxes, image_shape): # Collects outputs from all levels into a list. boxes = [] scores = [] for i in range(self._min_level, self._max_level + 1): batch_size = tf.shape(class_outputs[i])[0] # Applies score transformation and remove the implicit background class. scores_i = _apply_score_activation(class_outputs[i], self._num_classes, self._score_activation) # Box decoding. # The anchor boxes are shared for all data in a batch. # One stage detector only supports class agnostic box regression. anchor_boxes_i = tf.reshape(anchor_boxes[i], [batch_size, -1, 4]) box_outputs_i = tf.reshape(box_outputs[i], [batch_size, -1, 4]) boxes_i = box_utils.decode_boxes(box_outputs_i, anchor_boxes_i) # Box clipping. boxes_i = box_utils.clip_boxes(boxes_i, image_shape) boxes.append(boxes_i) scores.append(scores_i) boxes = tf.concat(boxes, axis=1) scores = tf.concat(scores, axis=1) boxes = tf.expand_dims(boxes, axis=2) (nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections) = self._generate_detections(boxes, scores) # Adds 1 to offset the background class which has index 0. nmsed_classes += 1 return nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections
def _decode_and_nms_fn(inputs, anchors): with tf.variable_scope("DecodeAndApplyNMS"): boxes_encoded = inputs[0] boxes_scores = inputs[1] boxes_decoded = box_utils.decode_boxes( boxes_encoded, anchors) boxes_resized = box_utils.resize_normalized_boxes( boxes_decoded, img_shape[0], img_shape[1]) boxes_clipped = box_utils.clip_to_img_boundaries( boxes_resized, image_shape=img_shape) boxes_probs = slim.softmax(boxes_scores) boxes_clipped_formatted = box_utils.convert_xyxy_to_yxyx_format( boxes_clipped) keep_boxes_ids = tf.image.non_max_suppression( boxes=boxes_clipped_formatted, scores=boxes_probs[:, 1], max_output_size=self.rpn_nms_num_samples, iou_threshold=self.rpn_nms_iou_th) boxes_out = tf.gather(boxes_clipped, keep_boxes_ids) probs_out = tf.gather(boxes_probs, keep_boxes_ids) return boxes_out, probs_out
def __call__(self, box_outputs, class_outputs, anchor_boxes, image_shape): # Collects outputs from all levels into a list. boxes = [] encoded_boxes = [] scores = [] for i in range(self._min_level, self._max_level + 1): _, feature_h, feature_w, num_predicted_corners = ( box_outputs[i].get_shape().as_list()) num_anchors_per_locations = num_predicted_corners // 4 num_classes = (class_outputs[i].get_shape().as_list()[-1] // num_anchors_per_locations) num_anchors = feature_h * feature_w * num_anchors_per_locations scores_i = tf.reshape(class_outputs[i], [-1, num_anchors, num_classes]) if self._apply_sigmoid: # Applies score transformation. scores_i = tf.sigmoid(scores_i) # Remove the implicit background class. scores_i = tf.slice(scores_i, [0, 0, 1], [-1, -1, -1]) # Box decoding. # The anchor boxes are shared for all data in a batch. # One stage detector only supports class agnostic box regression. anchor_boxes_i = tf.reshape(anchor_boxes[i], [-1, num_anchors, 4]) box_outputs_i = tf.reshape(box_outputs[i], [-1, num_anchors, 4]) encoded_boxes.append(box_outputs_i) boxes_i = box_utils.decode_boxes(box_outputs_i, anchor_boxes_i) # Box clipping. boxes_i = box_utils.clip_boxes(boxes_i, image_shape) boxes.append(boxes_i) scores.append(scores_i) boxes = tf.concat(boxes, axis=1) boxes = tf.expand_dims(boxes, axis=2) encoded_boxes = tf.concat(encoded_boxes, axis=1) scores = tf.concat(scores, axis=1) if not self._apply_nms: return { 'raw_boxes': boxes, 'raw_encoded_boxes': encoded_boxes, 'raw_scores': scores, } nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections = ( self._generate_detections(boxes, scores)) # Adds 1 to offset the background class which has index 0. nmsed_classes += 1 return { 'num_detections': valid_detections, 'detection_boxes': nmsed_boxes, 'detection_classes': nmsed_classes, 'detection_scores': nmsed_scores, }
def _box_outputs_to_rois(self, box_outputs, rois, correct_class, image_info, regression_weights): """Convert the box_outputs to be the new rois for the next cascade. Args: box_outputs: `tensor` with predicted bboxes in the most recent frcnn head. The predictions are relative to the anchors/rois, so we must convert them to x/y min/max to be used as rois in the following layer. rois: `tensor`, the rois used as input for frcnn head. correct_class: `tensor` of classes that the box should be predicted for. Used to filter the correct bbox prediction since they are done for all classes if `class_agnostic_bbox_pred` is not set to true. image_info: `list`, the height and width of the input image. regression_weights: `list`, weights used for l1 loss in bounding box regression. Returns: new_rois: rois to be used for the next frcnn layer in the cascade. """ if self._class_agnostic_bbox_pred: new_rois = box_outputs else: dtype = box_outputs.dtype batch_size, num_rois, num_class_specific_boxes = ( box_outputs.get_shape().as_list()) num_classes = num_class_specific_boxes // 4 box_outputs = tf.reshape(box_outputs, [batch_size, num_rois, num_classes, 4]) # correct_class is of shape [batch_size, num_rois]. # correct_class_one_hot has shape [batch_size, num_rois, num_classes, 4]. correct_class_one_hot = tf.tile( tf.expand_dims( tf.one_hot(correct_class, num_classes, dtype=dtype), -1), [1, 1, 1, 4]) new_rois = tf.reduce_sum(box_outputs * correct_class_one_hot, axis=2) new_rois = tf.cast(new_rois, tf.float32) # Before new_rois are predicting the relative center coords and # log scale offsets, so we need to run decode on them to get # the x/y min/max values needed for roi operations. # operations. new_rois = box_utils.decode_boxes(new_rois, rois, weights=regression_weights) new_rois = box_utils.clip_boxes(new_rois, image_info) return new_rois
def __call__(self, box_outputs, class_outputs, anchor_boxes): # Collects outputs from all levels into a list. boxes = [] scores = [] for i in range(self._min_level, self._max_level + 1): batch_size = tf.shape(class_outputs[i])[0] scores_i = _apply_score_activation(class_outputs[i], self._num_classes, self._score_activation) # The anchor boxes are shared for all data in a batch. # One stage detector only supports class agnostic box regression. anchor_boxes_i = tf.reshape(anchor_boxes[i], [batch_size, -1, 4]) box_outputs_i = tf.reshape(box_outputs[i], [batch_size, -1, 4]) boxes_i = box_utils.decode_boxes(box_outputs_i, anchor_boxes_i) boxes.append(boxes_i) scores.append(scores_i) boxes = tf.concat(boxes, axis=1) scores = tf.concat(scores, axis=1) boxes = tf.expand_dims(boxes, axis=2) (nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections) = self._generate_detections(boxes, scores) return nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections
def multilevel_propose_rois(rpn_boxes, rpn_scores, anchor_boxes, image_shape, rpn_pre_nms_top_k=2000, rpn_post_nms_top_k=1000, rpn_nms_threshold=0.7, rpn_score_threshold=0.0, rpn_min_size_threshold=0.0, decode_boxes=True, clip_boxes=True, use_batched_nms=False, apply_sigmoid_to_score=True): """Proposes RoIs given a group of candidates from different FPN levels. The following describes the steps: 1. For each individual level: a. Apply sigmoid transform if specified. b. Decode boxes if specified. c. Clip boxes if specified. d. Filter small boxes and those fall outside image if specified. e. Apply pre-NMS filtering including pre-NMS top k and score thresholding. f. Apply NMS. 2. Aggregate post-NMS boxes from each level. 3. Apply an overall top k to generate the final selected RoIs. Args: rpn_boxes: a dict with keys representing FPN levels and values representing box tenors of shape [batch_size, feature_h, feature_w, num_anchors * 4]. rpn_scores: a dict with keys representing FPN levels and values representing logit tensors of shape [batch_size, feature_h, feature_w, num_anchors]. anchor_boxes: a dict with keys representing FPN levels and values representing anchor box tensors of shape [batch_size, feature_h, feature_w, num_anchors * 4]. image_shape: a tensor of shape [batch_size, 2] where the last dimension are [height, width] of the scaled image. rpn_pre_nms_top_k: an integer of top scoring RPN proposals *per level* to keep before applying NMS. Default: 2000. rpn_post_nms_top_k: an integer of top scoring RPN proposals *in total* to keep after applying NMS. Default: 1000. rpn_nms_threshold: a float between 0 and 1 representing the IoU threshold used for NMS. If 0.0, no NMS is applied. Default: 0.7. rpn_score_threshold: a float between 0 and 1 representing the minimal box score to keep before applying NMS. This is often used as a pre-filtering step for better performance. If 0, no filtering is applied. Default: 0. rpn_min_size_threshold: a float representing the minimal box size in each side (w.r.t. the scaled image) to keep before applying NMS. This is often used as a pre-filtering step for better performance. If 0, no filtering is applied. Default: 0. decode_boxes: a boolean indicating whether `rpn_boxes` needs to be decoded using `anchor_boxes`. If False, use `rpn_boxes` directly and ignore `anchor_boxes`. Default: True. clip_boxes: a boolean indicating whether boxes are first clipped to the scaled image size before appliying NMS. If False, no clipping is applied and `image_shape` is ignored. Default: True. use_batched_nms: a boolean indicating whether NMS is applied in batch using `tf.image.combined_non_max_suppression`. Currently only available in CPU/GPU. Default: False. apply_sigmoid_to_score: a boolean indicating whether apply sigmoid to `rpn_scores` before applying NMS. Default: True. Returns: selected_rois: a tensor of shape [batch_size, rpn_post_nms_top_k, 1], representing the scores of the selected proposals. selected_roi_scores: a tensor of shape [batch_size, rpn_post_nms_top_k, 4], representing the box coordinates of the selected proposals w.r.t. the scaled image. """ with tf.name_scope('multilevel_propose_rois'): rois = [] roi_scores = [] for level in sorted(rpn_scores.keys()): with tf.name_scope('level_%d' % level): _, feature_h, feature_w, num_anchors_per_location = ( rpn_scores[level].get_shape().as_list()) num_boxes = feature_h * feature_w * num_anchors_per_location this_level_scores = tf.reshape(rpn_scores[level], [-1, num_boxes]) this_level_boxes = tf.reshape(rpn_boxes[level], [-1, num_boxes, 4]) this_level_anchors = tf.cast(tf.reshape( anchor_boxes[level], [-1, num_boxes, 4]), dtype=this_level_scores.dtype) if apply_sigmoid_to_score: this_level_scores = tf.sigmoid(this_level_scores) image_shape = tf.expand_dims(image_shape, axis=1) if decode_boxes: this_level_boxes = box_utils.decode_boxes( this_level_boxes, this_level_anchors) if clip_boxes: this_level_boxes = box_utils.clip_boxes( this_level_boxes, image_shape) if rpn_min_size_threshold > 0.0: this_level_boxes, this_level_scores = box_utils.filter_boxes( this_level_boxes, this_level_scores, image_shape, rpn_min_size_threshold) if rpn_nms_threshold > 0.0: this_level_pre_nms_top_k = min(num_boxes, rpn_pre_nms_top_k) if use_batched_nms: this_level_rois, this_level_roi_scores, _, _ = ( tf.image.combined_non_max_suppression( tf.expand_dims(this_level_boxes, axis=2), tf.expand_dims(this_level_scores, axis=-1), max_output_size_per_class= this_level_pre_nms_top_k, max_total_size=rpn_post_nms_top_k, iou_threshold=rpn_nms_threshold, score_threshold=rpn_score_threshold, pad_per_class=False, clip_boxes=False)) else: if rpn_score_threshold > 0.0: this_level_boxes, this_level_scores = ( box_utils.filter_boxes_by_scores( this_level_boxes, this_level_scores, rpn_score_threshold)) this_level_boxes, this_level_scores = box_utils.top_k_boxes( this_level_boxes, this_level_scores, k=this_level_pre_nms_top_k) this_level_roi_scores, this_level_rois = ( nms.sorted_non_max_suppression_padded( this_level_scores, this_level_boxes, max_output_size=rpn_post_nms_top_k, iou_threshold=rpn_nms_threshold)) else: this_level_rois, this_level_roi_scores = box_utils.top_k_boxes( this_level_rois, this_level_scores, k=rpn_post_nms_top_k) rois.append(this_level_rois) roi_scores.append(this_level_roi_scores) rois = tf.concat(rois, axis=1) roi_scores = tf.concat(roi_scores, axis=1) with tf.name_scope('top_k_rois'): _, num_valid_rois = roi_scores.get_shape().as_list() overall_top_k = min(num_valid_rois, rpn_post_nms_top_k) selected_rois, selected_roi_scores = box_utils.top_k_boxes( rois, roi_scores, k=overall_top_k) return selected_rois, selected_roi_scores
def __call__(self, box_outputs, class_outputs, anchor_boxes, image_shape, regression_weights=None, bbox_per_class=True, distill_class_outputs=None): """Generate final detections. Args: box_outputs: a tensor of shape of [batch_size, K, num_classes * 4] representing the class-specific box coordinates relative to anchors. class_outputs: a tensor of shape of [batch_size, K, num_classes] representing the class logits before applying score activation. anchor_boxes: a tensor of shape of [batch_size, K, 4] representing the corresponding anchor boxes w.r.t `box_outputs`. image_shape: a tensor of shape of [batch_size, 2] storing the image height and width w.r.t. the scaled image, i.e. the same image space as `box_outputs` and `anchor_boxes`. regression_weights: A list of four float numbers to scale coordinates. bbox_per_class: A `bool`. If True, perform per-class box regression. distill_class_outputs: a float tensor of shape of [batch_size, K, num_classes-1] representing the distilled class logits before applying score activation, without the background class. Returns: nmsed_boxes: `float` Tensor of shape [batch_size, max_total_size, 4] representing top detected boxes in [y1, x1, y2, x2]. nmsed_scores: `float` Tensor of shape [batch_size, max_total_size] representing sorted confidence scores for detected boxes. The values are between [0, 1]. nmsed_classes: `int` Tensor of shape [batch_size, max_total_size] representing classes for detected boxes. valid_detections: `int` Tensor of shape [batch_size] only the top `valid_detections` boxes are valid detections. """ class_outputs_shape = tf.shape(class_outputs) num_locations = class_outputs_shape[1] num_classes = class_outputs_shape[-1] if self._discard_background: # Removes the background class before softmax. class_outputs = tf.slice(class_outputs, [0, 0, 1], [-1, -1, -1]) class_outputs = tf.nn.softmax(class_outputs, axis=-1) if not self._discard_background: # Removes the background class. class_outputs = tf.slice(class_outputs, [0, 0, 1], [-1, -1, -1]) if self._feat_distill == 'double_branch': distill_class_outputs = tf.nn.softmax( distill_class_outputs, axis=-1) # [B, num_rois, num_classes] third_component = ( 1.0 - self._rare_mask ) * distill_class_outputs + self._rare_mask * class_outputs weighted_product = distill_class_outputs * class_outputs * third_component class_outputs = tf.pow(weighted_product, 1.0 / 3.0) if bbox_per_class: num_detections = num_locations * (num_classes - 1) box_outputs = tf.reshape(box_outputs, [-1, num_locations, num_classes, 4]) box_outputs = tf.slice(box_outputs, [0, 0, 1, 0], [-1, -1, -1, -1]) anchor_boxes = tf.tile(tf.expand_dims(anchor_boxes, axis=2), [1, 1, num_classes - 1, 1]) box_outputs = tf.reshape(box_outputs, [-1, num_detections, 4]) anchor_boxes = tf.reshape(anchor_boxes, [-1, num_detections, 4]) # Box decoding. if regression_weights is None: regression_weights = [10.0, 10.0, 5.0, 5.0] decoded_boxes = box_utils.decode_boxes(box_outputs, anchor_boxes, weights=regression_weights) # Box clipping decoded_boxes = box_utils.clip_boxes(decoded_boxes, image_shape) if bbox_per_class: decoded_boxes = tf.reshape(decoded_boxes, [-1, num_locations, num_classes - 1, 4]) else: decoded_boxes = tf.expand_dims(decoded_boxes, axis=2) if not self._apply_nms: return { 'raw_boxes': decoded_boxes, 'raw_scores': class_outputs, } nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections = ( self._generate_detections(decoded_boxes, class_outputs)) # Adds 1 to offset the background class which has index 0. nmsed_classes += 1 return { 'num_detections': valid_detections, 'detection_boxes': nmsed_boxes, 'detection_classes': nmsed_classes, 'detection_scores': nmsed_scores, }
def __call__(self, box_outputs, class_outputs, anchor_boxes, image_shape): """Generate final detections. Args: box_outputs: a tensor of shape of [batch_size, K, num_classes * 4] representing the class-specific box coordinates relative to anchors. class_outputs: a tensor of shape of [batch_size, K, num_classes] representing the class logits before applying score activiation. anchor_boxes: a tensor of shape of [batch_size, K, 4] representing the corresponding anchor boxes w.r.t `box_outputs`. image_shape: a tensor of shape of [batch_size, 2] storing the image height and width w.r.t. the scaled image, i.e. the same image space as `box_outputs` and `anchor_boxes`. Returns: nms_boxes: `float` Tensor of shape [batch_size, max_total_size, 4] representing top detected boxes in [y1, x1, y2, x2]. nms_scores: `float` Tensor of shape [batch_size, max_total_size] representing sorted confidence scores for detected boxes. The values are between [0, 1]. nms_classes: `int` Tensor of shape [batch_size, max_total_size] representing classes for detected boxes. valid_detections: `int` Tensor of shape [batch_size] only the top `valid_detections` boxes are valid detections. """ class_outputs = tf.nn.softmax(class_outputs, axis=-1) # Removes the background class. class_outputs_shape = tf.shape(class_outputs) batch_size = class_outputs_shape[0] num_locations = class_outputs_shape[1] num_classes = class_outputs_shape[-1] num_detections = num_locations * (num_classes - 1) class_outputs = tf.slice(class_outputs, [0, 0, 1], [-1, -1, -1]) box_outputs = tf.reshape( box_outputs, tf.stack([batch_size, num_locations, num_classes, 4], axis=-1)) box_outputs = tf.slice( box_outputs, [0, 0, 1, 0], [-1, -1, -1, -1]) anchor_boxes = tf.tile( tf.expand_dims(anchor_boxes, axis=2), [1, 1, num_classes - 1, 1]) box_outputs = tf.reshape( box_outputs, tf.stack([batch_size, num_detections, 4], axis=-1)) anchor_boxes = tf.reshape( anchor_boxes, tf.stack([batch_size, num_detections, 4], axis=-1)) # Box decoding. decoded_boxes = box_utils.decode_boxes( box_outputs, anchor_boxes, weights=[10.0, 10.0, 5.0, 5.0]) # Box clipping decoded_boxes = box_utils.clip_boxes(decoded_boxes, image_shape) decoded_boxes = tf.reshape( decoded_boxes, tf.stack([batch_size, num_locations, num_classes - 1, 4], axis=-1)) nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections = ( self._generate_detections(decoded_boxes, class_outputs)) # Adds 1 to offset the background class which has index 0. nmsed_classes += 1 return nmsed_boxes, nmsed_scores, nmsed_classes, valid_detections
def __call__(self, proposals, delta): return box_utils.decode_boxes(delta, proposals, self.weights)
def _decode_and_nms_fn(inputs): with tf.variable_scope("DecodeAndApplyNMS"): boxes_encoded = inputs[0] boxes_scores = inputs[1] rois = inputs[2] boxes_probs = slim.softmax(boxes_scores) boxes_classes = tf.argmax(boxes_probs, axis=1) # Do not include background prediction boxes_probs_red = tf.reduce_max(boxes_probs[..., 1:], axis=1) boxes_classes_one_hot = tf.cast( tf.one_hot(boxes_classes, depth=self.params.num_things_classes + 1), tf.bool) pad_num = tf.shape(boxes_classes)[0] boxes_encoded_per_class = tf.boolean_mask( boxes_encoded, boxes_classes_one_hot) boxes_encoded_per_class = tf.reshape( boxes_encoded_per_class, [-1, 4]) # Decode boxes boxes_decoded = box_utils.decode_boxes( boxes_encoded_per_class, rois, scale_factors=self.roi_encoder_scales) # Clip boxes to image boundaries boxes_resized = box_utils.resize_normalized_boxes( boxes_decoded, img_shape[0], img_shape[1]) boxes_clipped = box_utils.clip_to_img_boundaries( boxes_resized, image_shape=img_shape) # Find indices of boxes with score above the threshold and gather indices = tf.reshape( tf.where( tf.greater(boxes_probs_red, self.params.det_nms_score_th)), [-1]) boxes_clipped = tf.gather(boxes_clipped, indices) boxes_probs_red = tf.gather(boxes_probs_red, indices) # Subtract the background class from the predicted classes boxes_classes = tf.gather(boxes_classes, indices) - 1 boxes_clipped_formatted = box_utils.convert_xyxy_to_yxyx_format( boxes_clipped) keep_boxes_ids = tf.image.non_max_suppression( boxes=boxes_clipped_formatted, scores=boxes_probs_red, max_output_size=pad_num, iou_threshold=self.params.det_nms_iou_th) boxes_out = tf.gather(boxes_clipped, keep_boxes_ids) probs_out = tf.gather(boxes_probs_red, keep_boxes_ids) class_out = tf.gather(boxes_classes, keep_boxes_ids) boxes_pad, num_boxes = box_utils.pad_boxes_and_return_num( boxes_out, pad_num) probs_pad = tf.pad(probs_out, [[0, pad_num - num_boxes]]) class_pad = tf.pad(class_out, [[0, pad_num - num_boxes]]) boxes_pad = tf.reshape(boxes_pad, [pad_num, 4]) probs_pad = tf.reshape(probs_pad, [pad_num]) class_pad = tf.reshape(class_pad, [pad_num]) return boxes_pad, class_pad, probs_pad, num_boxes