Bug in rcnn.py - loss function

[meyerjo]

As described in #185 I get a similar problem. I fixed it by applying the boolean_mask cls_flatten not only to the bbox_flatten tensor but also the bbox_offset_targets_labeled. Thus, the elements not only get deleted from the one tensor but also from the other.

In luminoth/models/fasterrcnn/rcnn.py from line 378 `

        # We use the flatten cls_target_one_hot as boolean mask for the
        # bboxes.
        cls_flatten = tf.cast(tf.reshape(
            cls_target_one_hot, [-1]), tf.bool, 'cls_flatten_as_bool')

        bbox_offset_cleaned = tf.boolean_mask(
            bbox_flatten, cls_flatten, 'bbox_offset_cleaned')

        bbox_offsets_target_labeled_cleaned = tf.boolean_mask(
            bbox_offsets_target_labeled, cls_flatten, 'bbox_offsets_target_labeled_cleaned'
        )

        # Calculate the smooth l1 loss between the "cleaned" bboxes
        # offsets (that means, the useful results) and the labeled
        # targets.
        reg_loss_per_proposal = smooth_l1_loss(
            bbox_offset_cleaned, bbox_offsets_target_labeled_cleaned,
            sigma=self._l1_sigma
        )

`

This is the error I get beforehand and is fixed with snippet above: File "~/code/ext/luminoth/luminoth/models/fasterrcnn/fasterrcnn.py", line 192, in loss prediction_dict['classification_prediction'] File "~/code/ext/luminoth/luminoth/models/fasterrcnn/rcnn.py", line 391, in loss sigma=self._l1_sigma File "~/code/ext/luminoth/luminoth/utils/losses.py", line 22, in smooth_l1_loss diff = bbox_prediction - bbox_target File "~/.local/lib/python2.7/site-packages/tensorflow/python/ops/math_ops.py", line 850, in binary_op_wrapper return func(x, y, name=name) File "~/.local/lib/python2.7/site-packages/tensorflow/python/ops/gen_math_ops.py", line 8188, in sub "Sub", x=x, y=y, name=name) File "~/.local/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper op_def=op_def) File "~/.local/lib/python2.7/site-packages/tensorflow/python/util/deprecation.py", line 454, in new_func return func(*args, **kwargs) File "~/.local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 3155, in create_op op_def=op_def) File "~/.local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1717, in init self._traceback = tf_stack.extract_stack()

InvalidArgumentError (see above for traceback): Incompatible shapes: [27,4] vs. [28,4] [[Node: losses/RCNNLoss/sub_1 = Sub[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"](losses/RCNNLoss/bbox_offset_cleaned/GatherV2, losses/RCNNLoss/bbox_offsets_target_labeled/GatherV2)]] [[Node: losses/total_loss_1/_5777 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_11279_losses/total_loss_1", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

Do you fix this problem or shall I do a pull request?