search

yuantailing / ctw-baseline

Baseline methods for [CTW dataset](https://ctwdataset.github.io/)
MIT License
329 stars 88 forks source link

restore the weight of Inception_v4 #31

Open xgmiao opened 5 years ago

xgmiao commented 5 years ago

nvalidArgumentError (see above for traceback): Assign requires shapes of both tensors to match. lhs shape= [3,3,128,768] rhs shape= 5,5,128,768, use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](InceptionV4/AuxLogits/Conv2d_2a/weights, save/RestoreV2:9)]]

like: https://github.com/tensorflow/tensorflow/issues/18725

xgmiao commented 5 years ago

@yuantailing

yuantailing commented 5 years ago

What script are you executing? Where is your model from and where are your weights from?

xgmiao commented 5 years ago

@yuantailing the script is eval_image_classifier.py，get the model from https://ctwdataset.github.io/, together with the dataset

yuantailing commented 5 years ago

I think you'd better cd classification and run python3 eval.py inception_v4, it will call eval_image_classifier.py with correct arguments. It will fail if you run eval_image_classifier.py with wrong arguments.

xgmiao commented 5 years ago

@yuantailing thank you very much, I write a inference script to classify one image. solve the problem , it because the inception_v4 default input size is 299, and your trained model is 235. Thank you!
is the inception_v4 have the best performance? and the accuracy is ??

yuantailing commented 5 years ago

Results indicate that inception_v4 has the best performance even if input size is 235. Accuracy can be found in paper. (80.5%)