obj detection training with faster_rcnn_inception_v2_coco Argument must be a dense tensor: range(0, 3) - got shape [3], but wanted [].

[FalcoGer]

System information

• What is the top-level directory of the model you are using:
  TensorFlowModels\research\object_detection
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow):
  No, though PycocoAPI from here
  The problem also appeared without those modifications I made:

diff --git a/research/object_detection/core/box_predictor.py b/research/object_detection/core/box_predictor.py
index 6a13970..298a1dc 100644
--- a/research/object_detection/core/box_predictor.py
+++ b/research/object_detection/core/box_predictor.py
@@ -392,7 +392,7 @@ class MaskRCNNBoxPredictor(BoxPredictor):
         the proposals.
     """
     spatial_averaged_image_features = tf.reduce_mean(image_features, [1, 2],
-                                                     keep_dims=True,
+                                                     keepdims=True,
                                                      name='AvgPool')
     flattened_image_features = slim.flatten(spatial_averaged_image_features)
     if self._use_dropout:
diff --git a/research/object_detection/core/losses.py b/research/object_detection/core/losses.py
index 8bc044c..70959b6 100644
--- a/research/object_detection/core/losses.py
+++ b/research/object_detection/core/losses.py
@@ -311,7 +311,7 @@ class WeightedSoftmaxClassificationLoss(Loss):
     num_classes = prediction_tensor.get_shape().as_list()[-1]
     prediction_tensor = tf.divide(
         prediction_tensor, self._logit_scale, name='scale_logit')
-    per_row_cross_ent = (tf.nn.softmax_cross_entropy_with_logits(
+    per_row_cross_ent = (tf.nn.softmax_cross_entropy_with_logits_v2(
         labels=tf.reshape(target_tensor, [-1, num_classes]),
         logits=tf.reshape(prediction_tensor, [-1, num_classes])))
     return tf.reshape(per_row_cross_ent, tf.shape(weights)) * weights
diff --git a/research/object_detection/trainer.py b/research/object_detection/trainer.py
index cf3429a..196af61 100644
--- a/research/object_detection/trainer.py
+++ b/research/object_detection/trainer.py
@@ -225,7 +225,7 @@ def train(create_tensor_dict_fn, create_model_fn, train_config, master, task,

     # Place the global step on the device storing the variables.
     with tf.device(deploy_config.variables_device()):
-      global_step = slim.create_global_step()
+      global_step = tf.train.create_global_step()

     with tf.device(deploy_config.inputs_device()):
       input_queue = create_input_queue(

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
  Windows 7 x64
• TensorFlow installed from (source or binary):
  Installed with pip
• TensorFlow version (use command below):
  1.7.0
• Bazel version (if compiling from source):
• CUDA/cuDNN version:
  CPU version
• GPU model and memory:
  CPU version
• Exact command to reproduce:
  python train.py --pipeline_config_path="PATH_TO_MODELS/Models/faster_rcnn_inception_v2_coco/pipeline.config" --train_dir="PATH_TO_MODELS/Models/faster_rcnn_inception_v2_coco/train"

Describe the problem

trying to run the command above with tfrecords in the form of

tf_example = tf.train.Example(features=tf.train.Features(feature={
        'image/height': dataset_util.int64_feature(height),
        'image/width': dataset_util.int64_feature(width),
        'image/filename': dataset_util.bytes_feature(filename),
        'image/source_id': dataset_util.bytes_feature(filename),
        'image/encoded': dataset_util.bytes_feature(encoded_jpg),
        'image/format': dataset_util.bytes_feature(image_format),
        'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
        'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
        'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
        'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
        'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
        'image/object/class/label': dataset_util.int64_list_feature(classes),
    }))

and the faster_rcnn_inception_v2_coco model from 2018_01_28 creates the following errors:

Source code / logs

python : WARNING:tensorflow:From C:\Program Files\Python36\lib\site-packages\tensorflow\contrib\learn\python\learn\datasets\base.py:198: retry (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.
Instructions for updating:
Use the retry module or similar alternatives.
INFO:tensorflow:Scale of 0 disables regularizer.
INFO:tensorflow:Scale of 0 disables regularizer.
INFO:tensorflow:Scale of 0 disables regularizer.
INFO:tensorflow:Scale of 0 disables regularizer.
INFO:tensorflow:depth of additional conv before box predictor: 0
Traceback (most recent call last):
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 510, in _apply_op_helper
    preferred_dtype=default_dtype)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\ops.py", line 1040, in internal_convert_to_tensor
    ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\constant_op.py", line 235, in _constant_tensor_conversion_function
    return constant(v, dtype=dtype, name=name)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\constant_op.py", line 214, in constant
    value, dtype=dtype, shape=shape, verify_shape=verify_shape))
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\tensor_util.py", line 442, in make_tensor_proto
    _GetDenseDimensions(values)))
ValueError: Argument must be a dense tensor: range(0, 3) - got shape [3], but wanted [].
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 524, in _apply_op_helper
    values, as_ref=input_arg.is_ref).dtype.name
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\ops.py", line 1040, in internal_convert_to_tensor
    ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\constant_op.py", line 235, in _constant_tensor_conversion_function
    return constant(v, dtype=dtype, name=name)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\constant_op.py", line 214, in constant
    value, dtype=dtype, shape=shape, verify_shape=verify_shape))
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\tensor_util.py", line 442, in make_tensor_proto
    _GetDenseDimensions(values)))
ValueError: Argument must be a dense tensor: range(0, 3) - got shape [3], but wanted [].
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
  File "train.py", line 167, in <module>
    tf.app.run()
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\platform\app.py", line 126, in run
    _sys.exit(main(argv))
  File "train.py", line 163, in main
    worker_job_name, is_chief, FLAGS.train_dir)
  File "C:\temp\ObjDetection\TensorFlowModels\research\object_detection\trainer.py", line 255, in train
    train_config.optimizer)
  File "C:\temp\ObjDetection\TensorFlowModels\research\object_detection\builders\optimizer_builder.py", line 50, in build
    learning_rate = _create_learning_rate(config.learning_rate)
  File "C:\temp\ObjDetection\TensorFlowModels\research\object_detection\builders\optimizer_builder.py", line 109, in _create_learning_rate
    learning_rate_sequence, config.warmup)
  File "C:\temp\ObjDetection\TensorFlowModels\research\object_detection\utils\learning_schedules.py", line 169, in manual_stepping
    [0] * num_boundaries))
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\ops\array_ops.py", line 2650, in where
    return gen_math_ops.select(condition=condition, x=x, y=y, name=name)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\ops\gen_math_ops.py", line 7112, in select
    "Select", condition=condition, t=x, e=y, name=name)
  File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 528, in _apply_op_helper
    (input_name, err))
ValueError: Tried to convert 't' to a tensor and failed. Error: Argument must be a dense tensor: range(0, 3) - got shape [3], but wanted [].

Note: I've been using the detection api last when tensorflow version 1.4 was current. since updating i couldn't run ssd_mobilenet_v1 anymore, since the pretrained model checkpoint seems to missmatch the configuration. so I decided to try faster_rcnn and get this issue instead.

my pipeline config:

model {
  faster_rcnn {
    num_classes: 1
    image_resizer {
      keep_aspect_ratio_resizer {
        min_dimension: 600
        max_dimension: 1024
      }
    }
    feature_extractor {
      type: "faster_rcnn_resnet101"
      first_stage_features_stride: 16
    }
    first_stage_anchor_generator {
      grid_anchor_generator {
        height_stride: 16
        width_stride: 16
        scales: 0.25
        scales: 0.5
        scales: 1.0
        scales: 2.0
        aspect_ratios: 0.5
        aspect_ratios: 1.0
        aspect_ratios: 2.0
      }
    }
    first_stage_box_predictor_conv_hyperparams {
      op: CONV
      regularizer {
        l2_regularizer {
          weight: 0.0
        }
      }
      initializer {
        truncated_normal_initializer {
          stddev: 0.01
        }
      }
    }
    first_stage_nms_score_threshold: 0.0
    first_stage_nms_iou_threshold: 0.70
    first_stage_max_proposals: 100
    first_stage_localization_loss_weight: 2.0
    first_stage_objectness_loss_weight: 1.0
    initial_crop_size: 14
    maxpool_kernel_size: 2
    maxpool_stride: 2
    second_stage_box_predictor {
      mask_rcnn_box_predictor {
        fc_hyperparams {
          op: FC
          regularizer {
            l2_regularizer {
              weight: 0.0
            }
          }
          initializer {
            variance_scaling_initializer {
              factor: 1.0
              uniform: true
              mode: FAN_AVG
            }
          }
        }
        use_dropout: false
        dropout_keep_probability: 1.0
      }
    }
    second_stage_post_processing {
      batch_non_max_suppression {
        score_threshold: 0.30
        iou_threshold: 0.60
        max_detections_per_class: 100
        max_total_detections: 100
      }
      score_converter: SOFTMAX
    }
    second_stage_localization_loss_weight: 2.0
    second_stage_classification_loss_weight: 1.0
  }
}
train_config {
  batch_size: 1
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  optimizer {
    momentum_optimizer {
      learning_rate {
        manual_step_learning_rate {
          initial_learning_rate: 0.0003
# ref: https://github.com/tensorflow/models/issues/3794
#          schedule {
#            step: 0
#            learning_rate: 0.0003
#          }
          schedule {
            step: 900000
            learning_rate: 3.0e-05
          }
          schedule {
            step: 1200000
            learning_rate: 3.0e-06
          }
        }
      }
      momentum_optimizer_value: 0.90
    }
    use_moving_average: false
  }
  gradient_clipping_by_norm: 10.0
  fine_tune_checkpoint: "PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/train/model.ckpt"
  from_detection_checkpoint: true
}

# cd							cd C:/temp/ObjDetection/TensorFlowModels/research/object_detection
# train:						python train.py --pipeline_config_path="PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/pipeline.config" --train_dir="PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/train"
# eval:							python eval.py --pipeline_config_path="PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/pipeline.config" --eval_dir="PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/eval" --checkpoint_dir="PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/train"
# tb:							tensorboard --logdir="PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/"
# extract graph:				python export_inference_graph.py --input_type image_tensor --pipeline_config_path "PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/pipeline.config" --output_directory "PATH_TO_PROJECT/ExportGraph --trained_checkpoint_prefix "PATH_TO_PROJECT/Models/faster_rcnn_resnet101_coco/train/model.ckpt-REPLACETHIS"

train_input_reader {
  label_map_path: "PATH_TO_PROJECT/Data/labelMap.txt"
  shuffle: true
  tf_record_input_reader {
    input_path: "PATH_TO_PROJECT/Data/train.tfrecord"
  }
}
eval_config {
  num_examples: 8000
  max_evals: 10
  use_moving_averages: false
#  metrics_set: "coco_detection_metrics"
}
eval_input_reader {
  label_map_path: "PATH_TO_PROJECT/Data/labelMap.txt"
  shuffle: false
  num_readers: 1
  tf_record_input_reader {
    input_path: "PATH_TO_PROJECT/Data/val.tfrecord"
  }
}

PATH_TO_PROJECT of course is not really written there, but rather the path would be giving away what I'm working on, and I signed an NDA, so I replaced it here.

[FalcoGer]

I just uninstalled tensorflow 1.7.0 and installed 1.5.0, then rerun the protoc compilation and setup.py install of the tensorflow models
the problem persists.
the same problem is also present with faster_rcnn_resnet101_coco
the same problem is also present with rfcn_resnet101_coco

ssd runs fine except for #3922 but it seems to train decently enough

[elkalash]

I'm actually having the same issue here,
WARNING :tensorflow:from"path"is deprecated and will be removed in a future version
:Instructions for updating:
Use the retry module or similar alternatives.
Traceback )most recent call last);
File "src/run_webcap.py",line 32 in module

[FalcoGer]

@elkalash the depreciate warning is not the issue here.

[robieta]

Hi. Could you modify object_detection\utils\learning_schedules.py to print out global_step, boundaries, and num_boundaries and report the values?

[Lanbig]

Got the same issue.

ValueError: Tried to convert 't' to a tensor and failed. Error: Argument must be a dense tensor: range(0, 3) - got shape [3], but wanted [].

Found the solution : #3705 (comment)

[robieta]

That does indeed seem to be the case. Thanks a lot for finding and linking that.

[Edward1900]

I got the same problem in win10, how to solve it?

[FalcoGer]

@yafengwa
#3705 (comment)

[mikemwx]

I got the same error when tried to train faster_rcnn_resnet101 coco from model zoo.
I solved it by packing the list/range appeared in (maybe)
spatial_averaged_image_features = tf.reduce_mean(image_features, [1, 2], keep_dims=True, name='AvgPool')
with
tf.convert_to_tensor(list( """your range or list""" ),dtype=np.int64)