[MaskRCNN] handle empty forground in frcnn head.

examples/FasterRCNN/model_frcnn.py

@@ -144,6 +144,7 @@ def fastrcnn_losses(labels, label_logits, fg_boxes, fg_box_logits):
     fg_inds = tf.where(labels > 0)[:, 0]
     fg_labels = tf.gather(labels, fg_inds)
     num_fg = tf.size(fg_inds, out_type=tf.int64)
+    empty_fg = tf.equal(num_fg, 0)
     if int(fg_box_logits.shape[1]) > 1:
         indices = tf.stack(
             [tf.range(num_fg), fg_labels], axis=1)  # #fgx2
@@ -157,16 +158,18 @@ def fastrcnn_losses(labels, label_logits, fg_boxes, fg_box_logits):
         accuracy = tf.reduce_mean(correct, name='accuracy')
         fg_label_pred = tf.argmax(tf.gather(label_logits, fg_inds), axis=1)
         num_zero = tf.reduce_sum(tf.to_int64(tf.equal(fg_label_pred, 0)), name='num_zero')
-        false_negative = tf.truediv(num_zero, num_fg, name='false_negative')
-        fg_accuracy = tf.reduce_mean(
-            tf.gather(correct, fg_inds), name='fg_accuracy')
+        false_negative = tf.where(
+            empty_fg, 0., tf.truediv(num_zero, num_fg), name='false_negative')
+        fg_accuracy = tf.where(
+            empty_fg, 0., tf.reduce_mean(tf.gather(correct, fg_inds)), name='fg_accuracy')
 
     box_loss = tf.losses.huber_loss(
         fg_boxes, fg_box_logits, reduction=tf.losses.Reduction.SUM)
     box_loss = tf.truediv(
         box_loss, tf.to_float(tf.shape(labels)[0]), name='box_loss')
 
-    add_moving_summary(label_loss, box_loss, accuracy, fg_accuracy, false_negative)
+    add_moving_summary(label_loss, box_loss, accuracy,
+                       fg_accuracy, false_negative, tf.to_float(num_fg, name='num_fg_label'))
     return label_loss, box_loss
 
 

examples/FasterRCNN/train.py

@@ -141,12 +141,13 @@ class ResNetC4Model(DetectionModel):
                                      tf.constant(cfg.FRCNN.BBOX_REG_WEIGHTS, dtype=tf.float32))
 
         if is_training:
+            all_losses = []
             # rpn loss
-            rpn_label_loss, rpn_box_loss = rpn_losses(
-                anchors.gt_labels, anchors.encoded_gt_boxes(), rpn_label_logits, rpn_box_logits)
+            all_losses.extend(rpn_losses(
+                anchors.gt_labels, anchors.encoded_gt_boxes(), rpn_label_logits, rpn_box_logits))
 
             # fastrcnn loss
-            fastrcnn_label_loss, fastrcnn_box_loss = fastrcnn_head.losses()
+            all_losses.extend(fastrcnn_head.losses())
 
             if cfg.MODE_MASK:
                 # maskrcnn loss
@@ -161,18 +162,13 @@ class ResNetC4Model(DetectionModel):
                     proposals.fg_inds_wrt_gt, 14,
                     pad_border=False)  # nfg x 1x14x14
                 target_masks_for_fg = tf.squeeze(target_masks_for_fg, 1, 'sampled_fg_mask_targets')
-                mrcnn_loss = maskrcnn_loss(mask_logits, proposals.fg_labels(), target_masks_for_fg)
-            else:
-                mrcnn_loss = 0.0
+                all_losses.append(maskrcnn_loss(mask_logits, proposals.fg_labels(), target_masks_for_fg))
 
             wd_cost = regularize_cost(
                 '.*/W', l2_regularizer(cfg.TRAIN.WEIGHT_DECAY), name='wd_cost')
+            all_losses.append(wd_cost)
 
-            total_cost = tf.add_n([
-                rpn_label_loss, rpn_box_loss,
-                fastrcnn_label_loss, fastrcnn_box_loss,
-                mrcnn_loss, wd_cost], 'total_cost')
-
+            total_cost = tf.add_n(all_losses, 'total_cost')
             add_moving_summary(total_cost, wd_cost)
             return total_cost
         else:
@@ -272,11 +268,11 @@ class ResNetFPNModel(DetectionModel):
                                      tf.constant(cfg.FRCNN.BBOX_REG_WEIGHTS, dtype=tf.float32))
 
         if is_training:
-            # rpn loss:
-            rpn_label_loss, rpn_box_loss = multilevel_rpn_losses(
-                multilevel_anchors, multilevel_label_logits, multilevel_box_logits)
+            all_losses = []
+            all_losses.extend(multilevel_rpn_losses(
+                multilevel_anchors, multilevel_label_logits, multilevel_box_logits))
 
-            fastrcnn_label_loss, fastrcnn_box_loss = fastrcnn_head.losses()
+            all_losses.extend(fastrcnn_head.losses())
 
             if cfg.MODE_MASK:
                 # maskrcnn loss
@@ -293,17 +289,13 @@ class ResNetFPNModel(DetectionModel):
                     proposals.fg_inds_wrt_gt, 28,
                     pad_border=False)  # fg x 1x28x28
                 target_masks_for_fg = tf.squeeze(target_masks_for_fg, 1, 'sampled_fg_mask_targets')
-                mrcnn_loss = maskrcnn_loss(mask_logits, proposals.fg_labels(), target_masks_for_fg)
-            else:
-                mrcnn_loss = 0.0
+                all_losses.append(maskrcnn_loss(mask_logits, proposals.fg_labels(), target_masks_for_fg))
 
             wd_cost = regularize_cost(
                 '.*/W', l2_regularizer(cfg.TRAIN.WEIGHT_DECAY), name='wd_cost')
+            all_losses.append(wd_cost)
 
-            total_cost = tf.add_n([rpn_label_loss, rpn_box_loss,
-                                   fastrcnn_label_loss, fastrcnn_box_loss,
-                                   mrcnn_loss, wd_cost], 'total_cost')
-
+            total_cost = tf.add_n(all_losses, 'total_cost')
             add_moving_summary(total_cost, wd_cost)
             return total_cost
         else:

tensorpack/tfutils/collection.py

@@ -141,7 +141,7 @@ class CollectionGuard(object):
                     size_change.append((self._key_name(k), len(old_v), len(v)))
         if newly_created:
             logger.info(
-                "New collections created in {}: {}".format(
+                "New collections created in tower {}: {}".format(
                     self._name, ', '.join(newly_created)))
         if size_change:
             logger.info(