Internal change

PiperOrigin-RevId: 370677512

official/vision/beta/configs/maskrcnn.py

@@ -100,6 +100,10 @@ class DetectionHead(hyperparams.Config):
   use_separable_conv: bool = False
   num_fcs: int = 1
   fc_dims: int = 1024
+  class_agnostic_bbox_pred: bool = False  # Has to be True for Cascade RCNN.
+  # If additional IoUs are passed in 'cascade_iou_thresholds'
+  # then ensemble the class probabilities from all heads.
+  cascade_class_ensemble: bool = False
 
 
 @dataclasses.dataclass
@@ -125,6 +129,9 @@ class ROISampler(hyperparams.Config):
   foreground_iou_threshold: float = 0.5
   background_iou_high_threshold: float = 0.5
   background_iou_low_threshold: float = 0.0
+  # IoU thresholds for additional FRCNN heads in Cascade mode.
+  # `foreground_iou_threshold` is the first threshold.
+  cascade_iou_thresholds: Optional[List[float]] = None
 
 
 @dataclasses.dataclass
@@ -282,7 +289,6 @@ def fasterrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
           'task.train_data.is_training != None',
           'task.validation_data.is_training != None'
       ])
-
   return config
 
 
@@ -292,6 +298,68 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
   steps_per_epoch = 500
   coco_val_samples = 5000
 
+  config = cfg.ExperimentConfig(
+      runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
+      task=MaskRCNNTask(
+          init_checkpoint='gs://cloud-tpu-checkpoints/vision-2.0/resnet50_imagenet/ckpt-28080',
+          init_checkpoint_modules='backbone',
+          annotation_file=os.path.join(COCO_INPUT_PATH_BASE,
+                                       'instances_val2017.json'),
+          model=MaskRCNN(
+              num_classes=91, input_size=[1024, 1024, 3], include_mask=True),
+          losses=Losses(l2_weight_decay=0.00004),
+          train_data=DataConfig(
+              input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
+              is_training=True,
+              global_batch_size=64,
+              parser=Parser(
+                  aug_rand_hflip=True, aug_scale_min=0.8, aug_scale_max=1.25)),
+          validation_data=DataConfig(
+              input_path=os.path.join(COCO_INPUT_PATH_BASE, 'val*'),
+              is_training=False,
+              global_batch_size=8)),
+      trainer=cfg.TrainerConfig(
+          train_steps=22500,
+          validation_steps=coco_val_samples // 8,
+          validation_interval=steps_per_epoch,
+          steps_per_loop=steps_per_epoch,
+          summary_interval=steps_per_epoch,
+          checkpoint_interval=steps_per_epoch,
+          optimizer_config=optimization.OptimizationConfig({
+              'optimizer': {
+                  'type': 'sgd',
+                  'sgd': {
+                      'momentum': 0.9
+                  }
+              },
+              'learning_rate': {
+                  'type': 'stepwise',
+                  'stepwise': {
+                      'boundaries': [15000, 20000],
+                      'values': [0.12, 0.012, 0.0012],
+                  }
+              },
+              'warmup': {
+                  'type': 'linear',
+                  'linear': {
+                      'warmup_steps': 500,
+                      'warmup_learning_rate': 0.0067
+                  }
+              }
+          })),
+      restrictions=[
+          'task.train_data.is_training != None',
+          'task.validation_data.is_training != None'
+      ])
+  return config
+
+
+@exp_factory.register_config_factory('cascadercnn_resnetfpn_coco')
+def cascadercnn_resnetfpn_coco() -> cfg.ExperimentConfig:
+  """COCO object detection with Cascade R-CNN."""
+  steps_per_epoch = 500
+  coco_val_samples = 5000
+
   config = cfg.ExperimentConfig(
       runtime=cfg.RuntimeConfig(mixed_precision_dtype='bfloat16'),
       task=MaskRCNNTask(
@@ -302,7 +370,10 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
           model=MaskRCNN(
               num_classes=91,
               input_size=[1024, 1024, 3],
-              include_mask=True),
+              include_mask=True,
+              roi_sampler=ROISampler(cascade_iou_thresholds=[0.6, 0.7]),
+              detection_head=DetectionHead(
+                  class_agnostic_bbox_pred=True, cascade_class_ensemble=True)),
           losses=Losses(l2_weight_decay=0.00004),
           train_data=DataConfig(
               input_path=os.path.join(COCO_INPUT_PATH_BASE, 'train*'),
@@ -347,7 +418,6 @@ def maskrcnn_resnetfpn_coco() -> cfg.ExperimentConfig:
           'task.train_data.is_training != None',
           'task.validation_data.is_training != None'
       ])
-
   return config
 
 

official/vision/beta/losses/maskrcnn_losses.py

@@ -173,13 +173,21 @@ class FastrcnnClassLoss(object):
 class FastrcnnBoxLoss(object):
   """Fast R-CNN box regression loss function."""
 
-  def __init__(self, huber_loss_delta: float):
-    # The delta is typically around the mean value of regression target.
-    # for instances, the regression targets of 512x512 input with 6 anchors on
-    # P2-P6 pyramid is about [0.1, 0.1, 0.2, 0.2].
+  def __init__(self,
+               huber_loss_delta: float,
+               class_agnostic_bbox_pred: bool = False):
+    """Initiate Faster RCNN box loss.
 
+    Args:
+      huber_loss_delta: the delta is typically around the mean value of
+        regression target. for instances, the regression targets of 512x512
+        input with 6 anchors on P2-P6 pyramid is about [0.1, 0.1, 0.2, 0.2].
+      class_agnostic_bbox_pred: if True, class agnostic bounding box prediction
+        is performed.
+    """
     self._huber_loss = tf.keras.losses.Huber(
         delta=huber_loss_delta, reduction=tf.keras.losses.Reduction.SUM)
+    self._class_agnostic_bbox_pred = class_agnostic_bbox_pred
 
   def __call__(self, box_outputs, class_targets, box_targets):
     """Computes the box loss (Fast-RCNN branch) of Mask-RCNN.
@@ -207,32 +215,35 @@ class FastrcnnBoxLoss(object):
     """
     with tf.name_scope('fast_rcnn_loss'):
       class_targets = tf.cast(class_targets, dtype=tf.int32)
-
-      # Selects the box from `box_outputs` based on `class_targets`, with which
-      # the box has the maximum overlap.
-      (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])
-
-      box_indices = tf.reshape(
-          class_targets + tf.tile(
-              tf.expand_dims(
-                  tf.range(batch_size) * num_rois * num_classes, 1),
-              [1, num_rois]) + tf.tile(
-                  tf.expand_dims(tf.range(num_rois) * num_classes, 0),
-                  [batch_size, 1]), [-1])
-
-      box_outputs = tf.matmul(
-          tf.one_hot(
-              box_indices,
-              batch_size * num_rois * num_classes,
-              dtype=box_outputs.dtype), tf.reshape(box_outputs, [-1, 4]))
-      box_outputs = tf.reshape(box_outputs, [batch_size, -1, 4])
+      if not self._class_agnostic_bbox_pred:
+        box_outputs = self._assign_class_targets(box_outputs, class_targets)
 
       return self._fast_rcnn_box_loss(box_outputs, box_targets, class_targets)
 
+  def _assign_class_targets(self, box_outputs, class_targets):
+    """Selects the box from `box_outputs` based on `class_targets`, with which the box has the maximum overlap."""
+    (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])
+
+    box_indices = tf.reshape(
+        class_targets + tf.tile(
+            tf.expand_dims(tf.range(batch_size) * num_rois * num_classes, 1),
+            [1, num_rois]) + tf.tile(
+                tf.expand_dims(tf.range(num_rois) * num_classes, 0),
+                [batch_size, 1]), [-1])
+
+    box_outputs = tf.matmul(
+        tf.one_hot(
+            box_indices,
+            batch_size * num_rois * num_classes,
+            dtype=box_outputs.dtype), tf.reshape(box_outputs, [-1, 4]))
+    box_outputs = tf.reshape(box_outputs, [batch_size, -1, 4])
+
+    return box_outputs
+
   def _fast_rcnn_box_loss(self, box_outputs, box_targets, class_targets,
                           normalizer=1.0):
     """Computes box regression loss."""
@@ -299,4 +310,3 @@ class MaskrcnnLoss(object):
       # The loss is normalized by the number of 1's in weights and
       # + 0.01 is used to avoid division by zero.
       return mask_loss / (tf.reduce_sum(weights) + 0.01)
-

official/vision/beta/modeling/factory.py

@@ -109,6 +109,7 @@ def build_maskrcnn(
       use_separable_conv=detection_head_config.use_separable_conv,
       num_fcs=detection_head_config.num_fcs,
       fc_dims=detection_head_config.fc_dims,
+      class_agnostic_bbox_pred=detection_head_config.class_agnostic_bbox_pred,
       activation=norm_activation_config.activation,
       use_sync_bn=norm_activation_config.use_sync_bn,
       norm_momentum=norm_activation_config.norm_momentum,
@@ -131,6 +132,7 @@ def build_maskrcnn(
       test_num_proposals=roi_generator_config.test_num_proposals,
       use_batched_nms=roi_generator_config.use_batched_nms)
 
+  roi_sampler_cascade = []
   roi_sampler_obj = roi_sampler.ROISampler(
       mix_gt_boxes=roi_sampler_config.mix_gt_boxes,
       num_sampled_rois=roi_sampler_config.num_sampled_rois,
@@ -140,6 +142,18 @@ def build_maskrcnn(
           roi_sampler_config.background_iou_high_threshold),
       background_iou_low_threshold=(
           roi_sampler_config.background_iou_low_threshold))
+  roi_sampler_cascade.append(roi_sampler_obj)
+  # Initialize addtional roi simplers for cascade heads.
+  if roi_sampler_config.cascade_iou_thresholds:
+    for iou in roi_sampler_config.cascade_iou_thresholds:
+      roi_sampler_obj = roi_sampler.ROISampler(
+          mix_gt_boxes=False,
+          num_sampled_rois=roi_sampler_config.num_sampled_rois,
+          foreground_iou_threshold=iou,
+          background_iou_high_threshold=iou,
+          background_iou_low_threshold=0.0,
+          skip_subsampling=True)
+      roi_sampler_cascade.append(roi_sampler_obj)
 
   roi_aligner_obj = roi_aligner.MultilevelROIAligner(
       crop_size=roi_aligner_config.crop_size,
@@ -186,12 +200,14 @@ def build_maskrcnn(
       rpn_head=rpn_head,
       detection_head=detection_head,
       roi_generator=roi_generator_obj,
-      roi_sampler=roi_sampler_obj,
+      roi_sampler=roi_sampler_cascade,
       roi_aligner=roi_aligner_obj,
       detection_generator=detection_generator_obj,
       mask_head=mask_head,
       mask_sampler=mask_sampler_obj,
-      mask_roi_aligner=mask_roi_aligner_obj)
+      mask_roi_aligner=mask_roi_aligner_obj,
+      class_agnostic_bbox_pred=detection_head_config.class_agnostic_bbox_pred,
+      cascade_class_ensemble=detection_head_config.cascade_class_ensemble)
   return model
 
 

official/vision/beta/modeling/heads/instance_heads.py

@@ -33,6 +33,7 @@ class DetectionHead(tf.keras.layers.Layer):
       use_separable_conv: bool = False,
       num_fcs: int = 2,
       fc_dims: int = 1024,
+      class_agnostic_bbox_pred: bool = False,
       activation: str = 'relu',
       use_sync_bn: bool = False,
       norm_momentum: float = 0.99,
@@ -54,6 +55,8 @@ class DetectionHead(tf.keras.layers.Layer):
         the predictions.
       fc_dims: An `int` number that represents the number of dimension of the FC
         layers.
+      class_agnostic_bbox_pred: `bool`, indicating whether bboxes should be
+        predicted for every class or not.
       activation: A `str` that indicates which activation is used, e.g. 'relu',
         'swish', etc.
       use_sync_bn: A `bool` that indicates whether to use synchronized batch
@@ -73,6 +76,7 @@ class DetectionHead(tf.keras.layers.Layer):
         'use_separable_conv': use_separable_conv,
         'num_fcs': num_fcs,
         'fc_dims': fc_dims,
+        'class_agnostic_bbox_pred': class_agnostic_bbox_pred,
         'activation': activation,
         'use_sync_bn': use_sync_bn,
         'norm_momentum': norm_momentum,
@@ -155,8 +159,11 @@ class DetectionHead(tf.keras.layers.Layer):
         kernel_regularizer=self._config_dict['kernel_regularizer'],
         bias_regularizer=self._config_dict['bias_regularizer'],
         name='detection-scores')
+
+    num_box_outputs = (4 if self._config_dict['class_agnostic_bbox_pred'] else
+                       self._config_dict['num_classes'] * 4)
     self._box_regressor = tf.keras.layers.Dense(
-        units=self._config_dict['num_classes'] * 4,
+        units=num_box_outputs,
         kernel_initializer=tf.keras.initializers.RandomNormal(stddev=0.001),
         bias_initializer=tf.zeros_initializer(),
         kernel_regularizer=self._config_dict['kernel_regularizer'],

official/vision/beta/modeling/layers/detection_generator.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 """Contains definitions of generators to generate the final detections."""
-from typing import Optional, Mapping
+from typing import List, Optional, Mapping
 # Import libraries
 import tensorflow as tf
 
@@ -432,8 +432,13 @@ class DetectionGenerator(tf.keras.layers.Layer):
     }
     super(DetectionGenerator, self).__init__(**kwargs)
 
-  def __call__(self, raw_boxes: tf.Tensor, raw_scores: tf.Tensor,
-               anchor_boxes: tf.Tensor, image_shape: tf.Tensor):
+  def __call__(self,
+               raw_boxes: tf.Tensor,
+               raw_scores: tf.Tensor,
+               anchor_boxes: tf.Tensor,
+               image_shape: tf.Tensor,
+               regression_weights: Optional[List[float]] = None,
+               bbox_per_class: bool = True):
     """Generates final detections.
 
     Args:
@@ -446,6 +451,8 @@ class DetectionGenerator(tf.keras.layers.Layer):
       image_shape: A `tf.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.
 
     Returns:
       If `apply_nms` = True, the return is a dictionary with keys:
@@ -473,27 +480,32 @@ class DetectionGenerator(tf.keras.layers.Layer):
     batch_size = box_scores_shape[0]
     num_locations = box_scores_shape_list[1]
     num_classes = box_scores_shape_list[-1]
-    num_detections = num_locations * (num_classes - 1)
 
     box_scores = tf.slice(box_scores, [0, 0, 1], [-1, -1, -1])
-    raw_boxes = tf.reshape(raw_boxes,
-                           [batch_size, num_locations, num_classes, 4])
-    raw_boxes = tf.slice(raw_boxes, [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])
-    raw_boxes = tf.reshape(raw_boxes, [batch_size, num_detections, 4])
-    anchor_boxes = tf.reshape(anchor_boxes, [batch_size, num_detections, 4])
+
+    if bbox_per_class:
+      num_detections = num_locations * (num_classes - 1)
+      raw_boxes = tf.reshape(raw_boxes,
+                             [batch_size, num_locations, num_classes, 4])
+      raw_boxes = tf.slice(raw_boxes, [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])
+      raw_boxes = tf.reshape(raw_boxes, [batch_size, num_detections, 4])
+      anchor_boxes = tf.reshape(anchor_boxes, [batch_size, num_detections, 4])
 
     # Box decoding.
     decoded_boxes = box_ops.decode_boxes(
-        raw_boxes, anchor_boxes, weights=[10.0, 10.0, 5.0, 5.0])
+        raw_boxes, anchor_boxes, weights=regression_weights)
 
     # Box clipping
     decoded_boxes = box_ops.clip_boxes(
         decoded_boxes, tf.expand_dims(image_shape, axis=1))
 
-    decoded_boxes = tf.reshape(decoded_boxes,
-                               [batch_size, num_locations, num_classes - 1, 4])
+    if bbox_per_class:
+      decoded_boxes = tf.reshape(
+          decoded_boxes, [batch_size, num_locations, num_classes - 1, 4])
+    else:
+      decoded_boxes = tf.expand_dims(decoded_boxes, axis=2)
 
     if not self._config_dict['apply_nms']:
       return {

official/vision/beta/modeling/layers/roi_sampler.py

@@ -31,6 +31,7 @@ class ROISampler(tf.keras.layers.Layer):
                foreground_iou_threshold: float = 0.5,
                background_iou_high_threshold: float = 0.5,
                background_iou_low_threshold: float = 0,
+               skip_subsampling: bool = False,
                **kwargs):
     """Initializes a ROI sampler.
 
@@ -48,6 +49,9 @@ class ROISampler(tf.keras.layers.Layer):
       background_iou_low_threshold: A `float` that represents the IoU threshold
         for a box to be considered as negative (if overlap in
         [`background_iou_low_threshold`, `background_iou_high_threshold`])
+      skip_subsampling: a bool that determines if we want to skip the sampling
+        procedure than balances the fg/bg classes. Used for upper frcnn layers
+        in cascade RCNN.
       **kwargs: Additional keyword arguments passed to Layer.
     """
     self._config_dict = {
@@ -57,6 +61,7 @@ class ROISampler(tf.keras.layers.Layer):
         'foreground_iou_threshold': foreground_iou_threshold,
         'background_iou_high_threshold': background_iou_high_threshold,
         'background_iou_low_threshold': background_iou_low_threshold,
+        'skip_subsampling': skip_subsampling,
     }
 
     self._sim_calc = keras_cv.ops.IouSimilarity()
@@ -110,8 +115,8 @@ class ROISampler(tf.keras.layers.Layer):
         tensor, i.e.,
         gt_boxes[sampled_gt_indices[:, i]] = sampled_gt_boxes[:, i].
     """
+    gt_boxes = tf.cast(gt_boxes, dtype=boxes.dtype)
     if self._config_dict['mix_gt_boxes']:
-      gt_boxes = tf.cast(gt_boxes, dtype=boxes.dtype)
       boxes = tf.concat([boxes, gt_boxes], axis=1)
 
     boxes_invalid_mask = tf.less(
@@ -143,6 +148,10 @@ class ROISampler(tf.keras.layers.Layer):
         tf.squeeze(background_mask, -1), -tf.ones_like(matched_gt_indices),
         matched_gt_indices)
 
+    if self._config_dict['skip_subsampling']:
+      return (boxes, matched_gt_boxes, tf.squeeze(matched_gt_classes,
+                                                  axis=-1), matched_gt_indices)
+
     sampled_indices = self._sampler(
         positive_matches, negative_matches, ignored_matches)
 

official/vision/beta/modeling/maskrcnn_model.py

@@ -14,7 +14,7 @@
 
 """Mask R-CNN model."""
 
-from typing import Any, Mapping, Optional, Union
+from typing import Any, List, Mapping, Optional, Union
 
 # Import libraries
 import tensorflow as tf
@@ -32,12 +32,15 @@ class MaskRCNNModel(tf.keras.Model):
                rpn_head: tf.keras.layers.Layer,
                detection_head: tf.keras.layers.Layer,
                roi_generator: tf.keras.layers.Layer,
-               roi_sampler: tf.keras.layers.Layer,
+               roi_sampler: Union[tf.keras.layers.Layer,
+                                  List[tf.keras.layers.Layer]],
                roi_aligner: tf.keras.layers.Layer,
                detection_generator: tf.keras.layers.Layer,
                mask_head: Optional[tf.keras.layers.Layer] = None,
                mask_sampler: Optional[tf.keras.layers.Layer] = None,
                mask_roi_aligner: Optional[tf.keras.layers.Layer] = None,
+               class_agnostic_bbox_pred: bool = False,
+               cascade_class_ensemble: bool = False,
                **kwargs):
     """Initializes the Mask R-CNN model.
 
@@ -47,12 +50,17 @@ class MaskRCNNModel(tf.keras.Model):
       rpn_head: the RPN head.
       detection_head: the detection head.
       roi_generator: the ROI generator.
-      roi_sampler: the ROI sampler.
+      roi_sampler: a single ROI sampler or a list of ROI samplers for cascade
+        detection heads.
       roi_aligner: the ROI aligner.
       detection_generator: the detection generator.
       mask_head: the mask head.
       mask_sampler: the mask sampler.
       mask_roi_aligner: the ROI alginer for mask prediction.
+      class_agnostic_bbox_pred: if True, perform class agnostic bounding box
+        prediction. Needs to be `True` for Cascade RCNN models.
+      cascade_class_ensemble: if True, ensemble classification scores over
+        all detection heads.
       **kwargs: keyword arguments to be passed.
     """
     super(MaskRCNNModel, self).__init__(**kwargs)
@@ -68,13 +76,22 @@ class MaskRCNNModel(tf.keras.Model):
         'mask_head': mask_head,
         'mask_sampler': mask_sampler,
         'mask_roi_aligner': mask_roi_aligner,
+        'class_agnostic_bbox_pred': class_agnostic_bbox_pred,
+        'cascade_class_ensemble': cascade_class_ensemble,
     }
     self.backbone = backbone
     self.decoder = decoder
     self.rpn_head = rpn_head
     self.detection_head = detection_head
     self.roi_generator = roi_generator
-    self.roi_sampler = roi_sampler
+    if not isinstance(roi_sampler, (list, tuple)):
+      self.roi_sampler = [roi_sampler]
+    else:
+      self.roi_sampler = roi_sampler
+    if len(self.roi_sampler) > 1 and not class_agnostic_bbox_pred:
+      raise ValueError(
+          '`class_agnostic_bbox_pred` needs to be True if multiple detection heads are specified.'
+      )
     self.roi_aligner = roi_aligner
     self.detection_generator = detection_generator
     self._include_mask = mask_head is not None
@@ -85,6 +102,13 @@ class MaskRCNNModel(tf.keras.Model):
     if self._include_mask and mask_roi_aligner is None:
       raise ValueError('`mask_roi_aligner` is not provided in Mask R-CNN.')
     self.mask_roi_aligner = mask_roi_aligner
+    # Weights for the regression losses for each FRCNN layer.
+    # TODO(xianzhi): Make the weights configurable.
+    self._cascade_layer_to_weights = [
+        [10.0, 10.0, 5.0, 5.0],
+        [20.0, 20.0, 10.0, 10.0],
+        [30.0, 30.0, 15.0, 15.0],
+    ]
 
   def call(self,
            images: tf.Tensor,
@@ -110,44 +134,50 @@ class MaskRCNNModel(tf.keras.Model):
     })
 
     # Generate RoIs.
-    rois, _ = self.roi_generator(
-        rpn_boxes, rpn_scores, anchor_boxes, image_shape, training)
+    current_rois, _ = self.roi_generator(rpn_boxes, rpn_scores, anchor_boxes,
+                                         image_shape, training)
 
-    if training:
-      rois = tf.stop_gradient(rois)
+    next_rois = current_rois
+    all_class_outputs = []
+    for cascade_num in range(len(self.roi_sampler)):
+      # In cascade RCNN we want the higher layers to have different regression
+      # weights as the predicted deltas become smaller and smaller.
+      regression_weights = self._cascade_layer_to_weights[cascade_num]
+      current_rois = next_rois
 
-      rois, matched_gt_boxes, matched_gt_classes, matched_gt_indices = (
-          self.roi_sampler(rois, gt_boxes, gt_classes))
-      # Assign target for the 2nd stage classification.
-      box_targets = box_ops.encode_boxes(
-          matched_gt_boxes, rois, weights=[10.0, 10.0, 5.0, 5.0])
-      # If the target is background, the box target is set to all 0s.
-      box_targets = tf.where(
-          tf.tile(
-              tf.expand_dims(tf.equal(matched_gt_classes, 0), axis=-1),
-              [1, 1, 4]),
-          tf.zeros_like(box_targets),
-          box_targets)
-      model_outputs.update({
-          'class_targets': matched_gt_classes,
-          'box_targets': box_targets,
-      })
+      (class_outputs, box_outputs, model_outputs, matched_gt_boxes,
+       matched_gt_classes, matched_gt_indices,
+       current_rois) = self._run_frcnn_head(
+           features=features,
+           rois=current_rois,
+           gt_boxes=gt_boxes,
+           gt_classes=gt_classes,
+           training=training,
+           model_outputs=model_outputs,
+           layer_num=cascade_num,
+           regression_weights=regression_weights)
+      all_class_outputs.append(class_outputs)
 
-    # RoI align.
-    roi_features = self.roi_aligner(features, rois)
+      # Generate ROIs for the next cascade head if there is any.
+      if cascade_num < len(self.roi_sampler) - 1:
+        next_rois = box_ops.decode_boxes(
+            tf.cast(box_outputs, tf.float32),
+            current_rois,
+            weights=regression_weights)
+        next_rois = box_ops.clip_boxes(next_rois,
+                                       tf.expand_dims(image_shape, axis=1))
 
-    # Detection head.
-    raw_scores, raw_boxes = self.detection_head(roi_features)
+    if not training:
+      if self._config_dict['cascade_class_ensemble']:
+        class_outputs = tf.add_n(all_class_outputs) / len(all_class_outputs)
 
-    if training:
-      model_outputs.update({
-          'class_outputs': raw_scores,
-          'box_outputs': raw_boxes,
-      })
-    else:
-      # Post-processing.
       detections = self.detection_generator(
-          raw_boxes, raw_scores, rois, image_shape)
+          box_outputs,
+          class_outputs,
+          current_rois,
+          image_shape,
+          regression_weights,
+          bbox_per_class=(not self._config_dict['class_agnostic_bbox_pred']))
       model_outputs.update({
           'detection_boxes': detections['detection_boxes'],
           'detection_scores': detections['detection_scores'],
@@ -159,12 +189,9 @@ class MaskRCNNModel(tf.keras.Model):
       return model_outputs
 
     if training:
-      rois, roi_classes, roi_masks = self.mask_sampler(
-          rois,
-          matched_gt_boxes,
-          matched_gt_classes,
-          matched_gt_indices,
-          gt_masks)
+      current_rois, roi_classes, roi_masks = self.mask_sampler(
+          current_rois, matched_gt_boxes, matched_gt_classes,
+          matched_gt_indices, gt_masks)
       roi_masks = tf.stop_gradient(roi_masks)
 
       model_outputs.update({
@@ -172,11 +199,11 @@ class MaskRCNNModel(tf.keras.Model):
           'mask_targets': roi_masks,
       })
     else:
-      rois = model_outputs['detection_boxes']
+      current_rois = model_outputs['detection_boxes']
       roi_classes = model_outputs['detection_classes']
 
     # Mask RoI align.
-    mask_roi_features = self.mask_roi_aligner(features, rois)
+    mask_roi_features = self.mask_roi_aligner(features, current_rois)
 
     # Mask head.
     raw_masks = self.mask_head([mask_roi_features, roi_classes])
@@ -191,6 +218,80 @@ class MaskRCNNModel(tf.keras.Model):
       })
     return model_outputs
 
+  def _run_frcnn_head(self, features, rois, gt_boxes, gt_classes, training,
+                      model_outputs, layer_num, regression_weights):
+    """Runs the frcnn head that does both class and box prediction.
+
+    Args:
+      features: `list` of features from the feature extractor.
+      rois: `list` of current rois that will be used to predict bbox refinement
+        and classes from.
+      gt_boxes: a tensor with a shape of [batch_size, MAX_NUM_INSTANCES, 4].
+        This tensor might have paddings with a negative value.
+      gt_classes: [batch_size, MAX_INSTANCES] representing the groundtruth box
+        classes. It is padded with -1s to indicate the invalid classes.
+      training: `bool`, if model is training or being evaluated.
+      model_outputs: `dict`, used for storing outputs used for eval and losses.
+      layer_num: `int`, the current frcnn layer in the cascade.
+      regression_weights: `list`, weights used for l1 loss in bounding box
+        regression.
+
+    Returns:
+      class_outputs: Class predictions for rois.
+      box_outputs: Box predictions for rois. These are formatted for the
+        regression loss and need to be converted before being used as rois
+        in the next stage.
+      model_outputs: Updated dict with predictions used for losses and eval.
+      matched_gt_boxes: If `is_training` is true, then these give the gt box
+        location of its positive match.
+      matched_gt_classes: If `is_training` is true, then these give the gt class
+         of the predicted box.
+      matched_gt_boxes: If `is_training` is true, then these give the box
+        location of its positive match.
+      matched_gt_indices: If `is_training` is true, then gives the index of
+        the positive box match. Used for mask prediction.
+      rois: The sampled rois used for this layer.
+    """
+    # Only used during training.
+    matched_gt_boxes, matched_gt_classes, matched_gt_indices = (None, None,
+                                                                None)
+    if training:
+      rois = tf.stop_gradient(rois)
+
+      current_roi_sampler = self.roi_sampler[layer_num]
+      rois, matched_gt_boxes, matched_gt_classes, matched_gt_indices = (
+          current_roi_sampler(rois, gt_boxes, gt_classes))
+      # Create bounding box training targets.
+      box_targets = box_ops.encode_boxes(
+          matched_gt_boxes, rois, weights=regression_weights)
+      # If the target is background, the box target is set to all 0s.
+      box_targets = tf.where(
+          tf.tile(
+              tf.expand_dims(tf.equal(matched_gt_classes, 0), axis=-1),
+              [1, 1, 4]), tf.zeros_like(box_targets), box_targets)
+      model_outputs.update({
+          'class_targets_{}'.format(layer_num)
+          if layer_num else 'class_targets':
+              matched_gt_classes,
+          'box_targets_{}'.format(layer_num) if layer_num else 'box_targets':
+              box_targets,
+      })
+
+    # Get roi features.
+    roi_features = self.roi_aligner(features, rois)
+
+    # Run frcnn head to get class and bbox predictions.
+    class_outputs, box_outputs = self.detection_head(roi_features)
+
+    model_outputs.update({
+        'class_outputs_{}'.format(layer_num) if layer_num else 'class_outputs':
+            class_outputs,
+        'box_outputs_{}'.format(layer_num) if layer_num else 'box_outputs':
+            box_outputs,
+    })
+    return (class_outputs, box_outputs, model_outputs, matched_gt_boxes,
+            matched_gt_classes, matched_gt_indices, rois)
+
   @property
   def checkpoint_items(
       self) -> Mapping[str, Union[tf.keras.Model, tf.keras.layers.Layer]]:

official/vision/beta/modeling/maskrcnn_model_test.py

@@ -125,17 +125,30 @@ class MaskRCNNModelTest(parameterized.TestCase, tf.test.TestCase):
     self.assertEqual(expected_num_params, model.count_params())
 
   @parameterized.parameters(
-      (False, False,),
-      (False, True,),
-      (True, False,),
-      (True, True,),
+      (False, False, False),
+      (False, True, False),
+      (False, False, True),
+      (False, True, True),
+      (True, False, False),
+      (True, True, False),
+      (True, False, True),
+      (True, True, True),
   )
-  def test_forward(self, include_mask, training):
+  def test_forward(self, include_mask, training, use_cascade_heads):
     num_classes = 3
     min_level = 3
     max_level = 4
     num_scales = 3
     aspect_ratios = [1.0]
+    if use_cascade_heads:
+      cascade_iou_thresholds = [0.6]
+      class_agnostic_bbox_pred = True
+      cascade_class_ensemble = True
+    else:
+      cascade_iou_thresholds = None
+      class_agnostic_bbox_pred = False
+      cascade_class_ensemble = False
+
     image_size = (256, 256)
     images = np.random.rand(2, image_size[0], image_size[1], 3)
     image_shape = np.array([[224, 100], [100, 224]])
@@ -159,9 +172,23 @@ class MaskRCNNModelTest(parameterized.TestCase, tf.test.TestCase):
         max_level=max_level,
         num_anchors_per_location=num_anchors_per_location)
     detection_head = instance_heads.DetectionHead(
-        num_classes=num_classes)
+        num_classes=num_classes,
+        class_agnostic_bbox_pred=class_agnostic_bbox_pred)
     roi_generator_obj = roi_generator.MultilevelROIGenerator()
+
+    roi_sampler_cascade = []
     roi_sampler_obj = roi_sampler.ROISampler()
+    roi_sampler_cascade.append(roi_sampler_obj)
+    if cascade_iou_thresholds:
+      for iou in cascade_iou_thresholds:
+        roi_sampler_obj = roi_sampler.ROISampler(
+            mix_gt_boxes=False,
+            foreground_iou_threshold=iou,
+            background_iou_high_threshold=iou,
+            background_iou_low_threshold=0.0,
+            skip_subsampling=True)
+        roi_sampler_cascade.append(roi_sampler_obj)
+
     roi_aligner_obj = roi_aligner.MultilevelROIAligner()
     detection_generator_obj = detection_generator.DetectionGenerator()
     if include_mask:
@@ -180,12 +207,14 @@ class MaskRCNNModelTest(parameterized.TestCase, tf.test.TestCase):
         rpn_head,
         detection_head,
         roi_generator_obj,
-        roi_sampler_obj,
+        roi_sampler_cascade,
         roi_aligner_obj,
         detection_generator_obj,
         mask_head,
         mask_sampler_obj,
-        mask_roi_aligner_obj)
+        mask_roi_aligner_obj,
+        class_agnostic_bbox_pred=class_agnostic_bbox_pred,
+        cascade_class_ensemble=cascade_class_ensemble)
 
     gt_boxes = np.array(
         [[[10, 10, 15, 15], [2.5, 2.5, 7.5, 7.5], [-1, -1, -1, -1]],

official/vision/beta/tasks/maskrcnn.py

@@ -161,6 +161,7 @@ class MaskRCNNTask(base_task.Task):
                    aux_losses: Optional[Any] = None):
     """Build Mask R-CNN losses."""
     params = self.task_config
+    cascade_ious = params.model.roi_sampler.cascade_iou_thresholds
 
     rpn_score_loss_fn = maskrcnn_losses.RpnScoreLoss(
         tf.shape(outputs['box_outputs'])[1])
@@ -175,15 +176,32 @@ class MaskRCNNTask(base_task.Task):
 
     frcnn_cls_loss_fn = maskrcnn_losses.FastrcnnClassLoss()
     frcnn_box_loss_fn = maskrcnn_losses.FastrcnnBoxLoss(
-        params.losses.frcnn_huber_loss_delta)
-    frcnn_cls_loss = tf.reduce_mean(
-        frcnn_cls_loss_fn(
-            outputs['class_outputs'], outputs['class_targets']))
-    frcnn_box_loss = tf.reduce_mean(
-        frcnn_box_loss_fn(
-            outputs['box_outputs'],
-            outputs['class_targets'],
-            outputs['box_targets']))
+        params.losses.frcnn_huber_loss_delta,
+        params.model.detection_head.class_agnostic_bbox_pred)
+
+    # Final cls/box losses are computed as an average of all detection heads.
+    frcnn_cls_loss = 0.0
+    frcnn_box_loss = 0.0
+    num_det_heads = 1 if cascade_ious is None else 1 + len(cascade_ious)
+    for cas_num in range(num_det_heads):
+      frcnn_cls_loss_i = tf.reduce_mean(
+          frcnn_cls_loss_fn(
+              outputs['class_outputs_{}'
+                      .format(cas_num) if cas_num else 'class_outputs'],
+              outputs['class_targets_{}'
+                      .format(cas_num) if cas_num else 'class_targets']))
+      frcnn_box_loss_i = tf.reduce_mean(
+          frcnn_box_loss_fn(
+              outputs['box_outputs_{}'.format(cas_num
+                                             ) if cas_num else 'box_outputs'],
+              outputs['class_targets_{}'
+                      .format(cas_num) if cas_num else 'class_targets'],
+              outputs['box_targets_{}'.format(cas_num
+                                             ) if cas_num else 'box_targets']))
+      frcnn_cls_loss += frcnn_cls_loss_i
+      frcnn_box_loss += frcnn_box_loss_i
+    frcnn_cls_loss /= num_det_heads
+    frcnn_box_loss /= num_det_heads
 
     if params.model.include_mask:
       mask_loss_fn = maskrcnn_losses.MaskrcnnLoss()