7

我正在尝试根据此处提供的官方 MaskRCNN 模型训练 Mask RCNN 模型：tensorflow/models。

以下是我遵循的步骤：

  1. 创建了一个用于训练和验证的 tfrecord。我检查了 tfrecords 的编码和解码，它工作正常。
  2. 设置配置文件如下： 
    # my_maskrcnn.yaml
train:
    train_file_pattern: "data/<dataset_name>/train/tfrecords/train.tfrecord-*"
    batch_size: 2
eval:
    eval_file_pattern: "data/<data_set_name>/val/tfrecords/val.tfrecord-*"
    batch_size: 2
predict:
    batch_size: 2
architecture:
    num_classes: 2
maskrcnn_parser:
    output_size: [512, 512]
  3. 将模型/官方放入 PYTHONPATH。
  4. 按照文档中的建议运行单个 GPU 的命令： 
    python path/to/models/official/vision/detection/main.py \
    --strategy_type=one_device \
    --model_dir=models_mask_rcnn \
    --mode=train \
    --config_file="mymaskrcnn.yaml" \
    --model=mask_rcnn

但我收到以下错误：

Traceback (most recent call last):
File ".../models/official/vision/detection/main.py", line 255, in <module>
    app.run(main)
File ".../lib/python3.7/site-packages/absl/app.py", line 299, in run
    _run_main(main, args)
File ".../lib/python3.7/site-packages/absl/app.py", line 250, in _run_main
    sys.exit(main(argv))
File ".../models/official/vision/detection/main.py", line 250, in main
    run()
File ".../models/official/vision/detection/main.py", line 244, in run
    callbacks=callbacks)
File ".../models/official/vision/detection/main.py", line 130, in run_executor
    save_config=True)
File ".../models/official/modeling/training/distributed_executor.py", line 482, in train
    tf.convert_to_tensor(num_steps, dtype=tf.int32))
File ".../lib/python3.7/site-packages/tensorflow/python/eager/def_function.py", line 688, in __call__
    result = self._call(*args, **kwds)
File ".../lib/python3.7/site-packages/tensorflow/python/eager/def_function.py", line 741, in _call
    return self._stateless_fn(*args, **kwds)
File ".../lib/python3.7/site-packages/tensorflow/python/eager/function.py", line 2407, in __call__
    return graph_function._filtered_call(args, kwargs)  # pylint: disable=protected-access
File ".../lib/python3.7/site-packages/tensorflow/python/eager/function.py", line 1655, in _filtered_call
    self.captured_inputs)
File ".../lib/python3.7/site-packages/tensorflow/python/eager/function.py", line 1732, in _call_flat
    ctx, args, cancellation_manager=cancellation_manager))
File ".../lib/python3.7/site-packages/tensorflow/python/eager/function.py", line 598, in call
    ctx=ctx)
File ".../lib/python3.7/site-packages/tensorflow/python/eager/execute.py", line 60, in quick_execute
    inputs, attrs, num_outputs)
tensorflow.python.framework.errors_impl.InvalidArgumentError: 2 root error(s) found.
(0) Invalid argument:  indices[1,63] = [1, -1] does not index into param shape [2,100,112,112]
        [[{{node while/body/_241/while/maskrcnn/tf_op_layer_sample_and_crop_foreground_masks/GatherNd_4/sample_and_crop_foreground_masks/GatherNd_4}}]]
(1) Invalid argument:  indices[1,63] = [1, -1] does not index into param shape [2,100,112,112]
        [[{{node while/body/_241/while/maskrcnn/tf_op_layer_sample_and_crop_foreground_masks/GatherNd_4/sample_and_crop_foreground_masks/GatherNd_4}}]]
        [[while/body/_241/while/AddN_40/_3885]]
0 successful operations.
1 derived errors ignored. [Op:__inference_train_step_89096]

Function call stack:
train_step -> train_step

我哪里错了？

4

0 回答 0