fail to train for unet 3D

[jinxsfe]

Hi, whether adjust UNET structure for 4 depth, instead of 3 depth, and training show an error, for weight names,

ValueError Traceback (most recent call last)
in <cell line: 18>()
16 start = time.time()
17 # Start Training
---> 18 model.train(epochs=number_of_epochs,
19 batch_size=batch_size,
20 train_generator=train_generator,

1 frames
/usr/local/lib/python3.10/dist-packages/keras/src/callbacks/model_checkpoint.py in init(self, filepath, monitor, verbose, save_best_only, save_weights_only, mode, save_freq, initial_value_threshold)
181 if save_weights_only:
182 if not self.filepath.endswith(".weights.h5"):
--> 183 raise ValueError(
184 "When using save_weights_only=True in ModelCheckpoint"
185 ", the filepath provided must end in .weights.h5 "

ValueError: When using save_weights_only=True in ModelCheckpoint, the filepath provided must end in .weights.h5 (Keras weights format). Received: filepath=/content/gdrive/MyDrive/cryo/cryo-data_processing_volume/model/gaussian4layer_320_50_0.00014/ckpt/gaussian4layer_320_50_0.00014.hdf5

I tried to change if save_best_ckpt_only:
ckpt_name = ckpt_dir + '/' + model_name + '.hdf5'
else:
ckpt_name = ckpt_dir + '/' + model_name + 'epoch{epoch:02d}val_loss{val_loss:.4f}.hdf5' to if save_best_ckpt_only:
ckpt_name = ckpt_dir + '/' + model_name + '.weights.h5'
else:
ckpt_name = ckpt_dir + '/' + model_name + 'epoch{epoch:02d}val_loss{val_loss:.4f}.weighths.h5' or keras corresponding to the save_weight"_only = True and save best only =True requirement but error output.

I even go model check point to set these two parameter equal true and false, but also cause the same issue

[jinxsfe]

@esgomezm @paxcalpt

[jinxsfe]

I kept the original structure for your notebook but cause the same error

[jinxsfe]

when I change back to ckpt_name = ckpt_dir + '/' + model_name + '.hdf5'
else:
ckpt_name = ckpt_dir + '/' + model_name + 'epoch{epoch:02d}val_loss{val_loss:.4f}.hdf5'.

and turn these off, it show the error

if I change it to keras, also show the same problem for

[jinxsfe]

sorry I forgot to input .hdf5 but error still occur

[IvanHCenalmor]

Hi @jinxsfe,

Not 100% sure but the problem seems to be with the name of the weights. It's looking for weights ending on .keras, this might be a functionality of the latest Keras and TensorFlow version.

On a new version of the U-Net 3D notebook, these versions are fixed to oldest ones so this issue might not happen.

While it's under testing the new version of the notebook it's on a different branch. You can find the updated notebook here:
U-Net_3D_ZeroCostDL4Mic.ipynb Open In Colab

Please feel free to test it and let us know if you encounter any further issues or if this resolves the problem.

Thanks again for your valuable feedback!

Iván

[jinxsfe]

@IvanHCenalmor , seem still missing package, even I run "!pip install imagecodecs
import imagecodecs, the error still show there

[jinxsfe]

@cleterrier @paxcalpt