1

I followed the following steps：

1.CityScapes data set preparation

2.Generate TFRecords of CityScapes

3.Download the pre-training model

4.Run official instruction

python deeplab/train.py \
--logtostderr \
--training_number_of_steps=1000 \
--train_split="train" \
--model_variant="xception_65" \
--atrous_rates=6 \
--atrous_rates=12 \
--atrous_rates=18 \
--output_stride=16 \
--decoder_output_stride=4 \
--train_crop_size=513 \
--train_crop_size=513 \
--train_batch_size=1 \
--dataset="cityscapes" \
--tf_initial_checkpoint='/root/newP/official_tf/models-master/research/deeplab/backbone/deeplabv3_cityscapes_train/model.ckpt' \
--train_logdir='/root/newP/official_tf/models-master/research/deeplab/exp/train_on_train_set/train' \
--dataset_dir='/root/dataset/cityscapesScripts/tfrecord'

An error occurred while training deeplabv3++ using CityScapes dataset

“data split name train not recognized”.

I found the problem after debugging: "train" no longer exit in

"_CITYSCAPES_INFORMATION.splits_to_sizes". Content in the code：

 _CITYSCAPES_INFORMATION = DatasetDescriptor(
    splits_to_sizes={'train_fine': 2975,
                     'train_coarse': 22973,
                     'trainval_fine': 3475,
                     'trainval_coarse': 23473,
                     'val_fine': 500,
                     'test_fine': 1525},
    num_classes=19,
    ignore_label=255,
)

I tried several others "train_fine","train_coarse".A new error occurred:

"Total size of new array must be unchanged for image_pooling/weights lh_shape: [(1, 1, 2048, 256)], rh_shape: [(1, 1, 320, 256)]".

May I ask what modifications I should do?

Rishabh Sahrawat
  • 2,437
  • 1
  • 15
  • 32
Andrew
  • 11
  • 1

1 Answers1

0

I found that the latest version of the pretraining model had a problem, and I could run it directly when I was not using the pretraining model. [https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md][1]

Andrew
  • 11
  • 1