Models: [deeplab v3+] I want to use the model on my own dataset, which just has 8 classes

[smmdream]

InvalidArgumentError (see above for traceback): Assign requires shapes of both tensors to match. lhs shape= [8] rhs shape= [21]

 [[Node: save/Assign_113 = Assign[T=DT_FLOAT, _class=["loc:@logits/semantic/biases"], use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](logits/semantic/biases, save/RestoreV2_113)]]

The class number of my dataset is 8, and the model which trains on the VOC2012 has 21 classes, I don't know how to train on my own dataset, could you help me?

[95xueqian]

https://github.com/tensorflow/models/issues/3730 may help you

[smmdream]

@95xueqian Thank you very much, I do as the answer say, but it doesn't solve my question, at last, I delete the file named checkpoint in ./datasets/segmentation_dataset/exp/train_on_trainval_set/train/ , then solve my problem. Thank you very much.

[urgonguyen]

@smmdream Did you complete training your own dataset? I need to train deeplabv3+ on my data( 1000 images and 1000 augmented images, 17 label), How i can divide my dataset to effectively train( train, val, test), Could you give me some advices?Thanks

[yhliang2018]

As the problem is solved, I will close the issue. @urgonguyen You can contact @95xueqian for your question, or open another issue if you need.

[shaunster0]

@smmdream thanks for this, I had this problem too and ur advice about deleting existing files in /datasets/segmentation_dataset/exp/train_on_trainval_set/train/ worked

[kayvonkhosrowpour]

@smmdream What files specifically did you delete? I am having a similar error when I try to run a modified version of vis.py.
Edit:
Resolved issue: https://github.com/tensorflow/models/issues/6559#issuecomment-482671438

[beyou923]

@95xueqian Thank you very much, I do as the answer say, but it doesn't solve my question, at last, I delete the file named checkpoint in ./datasets/segmentation_dataset/exp/train_on_trainval_set/train/ , then solve my problem. Thank you very much.

hello, I deleted the checkpoint file according to your method, run evel.py stuck here:
INFO:tensorflow:Waiting for new checkpoint at ./datasets/ground_truth/exp/train_on_train_set/train