EfficientNetB3 meets error when load pretrained weights

[王嘉楠]

Thanks a lot! I can use EfficientNetB0 perfectly now. But when I change the base model to EfficientNetB3, an error raises:

Traceback (most recent call last):
  File "/data1/jiananwang/anaconda3/envs/fashion/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1628, in _create_c_op
    c_op = c_api.TF_FinishOperation(op_desc)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimension 0 in both shapes must be equal, but are 1 and 1536. Shapes are [1,1,384,1535] and [1536,384,1,1]. for 'Assign_489' (op: 'Assign') with input shapes: [1,1,384,1535], [1536,384,1,1].

I also tried EfficientNetB1, EfficientNetB2, similar error. Could you check it?

[Somshubra Majumdar]

Good catch. Could you pull from master and see if the latest fix 15e784a corrects this ?

[王嘉楠]

@titu1994 Yes, It works now. Thanks a lot!