ValueError: File format not supported: filepath=/root/.cache/doctr/models/parseq-24cf693e/weights. Keras 3 only supports V3 .keras and .weights.h5 files, or legacy V1/V2 .h5 files.

[abhijeetsourav]

Bug description

when i try to fine tune a pretrained parseq model on a custom dataset, I got the following error
How to fine tune a model on a custom dataset?
I want to generate a .pb file for running the model on edge devices?
So, I tried to save the model using model.save(<path/to/save>.keras)......
I don't know how to proceed further,
please help me to get the model file for running the parseq optimized model on edge devices, like android.....

Code snippet to reproduce the bug

!python3 references/recognition/train_tensorflow.py parseq --train_path <path/to/train_dataset> --val_path <path/to/eval_dataset> --epochs 1 --pretrained

Error traceback

Traceback (most recent call last):
  File "/content/doctr/references/recognition/train_tensorflow.py", line 447, in <module>
    main(args)
  File "/content/doctr/references/recognition/train_tensorflow.py", line 180, in main
    model = recognition.__dict__[args.arch](
  File "/content/doctr/doctr/models/recognition/parseq/tensorflow.py", line 505, in parseq
    return _parseq(
  File "/content/doctr/doctr/models/recognition/parseq/tensorflow.py", line 481, in _parseq
    load_pretrained_params(model, default_cfgs[arch]["url"])
  File "/content/doctr/doctr/models/utils/tensorflow.py", line 73, in load_pretrained_params
    model.load_weights(f"{params_path}{os.sep}{internal_name}")


  File "/usr/local/lib/python3.10/dist-packages/keras/src/utils/traceback_utils.py", line 122, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/usr/local/lib/python3.10/dist-packages/keras/src/saving/saving_api.py", line 268, in load_weights
    raise ValueError(
ValueError: File format not supported: filepath=/root/.cache/doctr/models/parseq-24cf693e/weights. Keras 3 only supports V3 `.keras` and `.weights.h5` files, or legacy V1/V2 `.h5` files.

Environment

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1725892688.189874 9133 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1725892688.197677 9133 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1725892688.197925 9133 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
DocTR version: 0.9.1a0
TensorFlow version: 2.17.0
PyTorch version: 2.4.0+cu121 (torchvision 0.19.0+cu121)
OpenCV version: 4.10.0
OS: Ubuntu 22.04.3 LTS
Python version: 3.10.12
Is CUDA available (TensorFlow): Yes
Is CUDA available (PyTorch): Yes
CUDA runtime version: 12.2.140
GPU models and configuration: GPU 0: Tesla T4
Nvidia driver version: 535.104.05
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.6
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.6

Deep Learning backend

is_tf_available: True
is_torch_available: True

[felixdittrich92]

Hi @abhijeetsourav 👋,
Ref.: #1461
We don't support Keras v3 atm please downgrade to

doctr/pyproject.toml

Line 57 in 9045dcf

"tensorflow>=2.11.0,<2.16.0",

and Keras v2 :)

[abhijeetsourav]

@felixdittrich92
Thanks for the quick reply...
please tell me how to convert the saved weights to .pb file for use in edge devices like android....
more specifically, I want to convert parseq model to run on android device....

[felixdittrich92]

You can try:

import tf 
from doctr.models import parseq
model = parseq(pretrained=True, exportable=True)
tf.saved_model.save(model, "parseq.pb")

Our Focus is more on Onnx here you can check out:
https://github.com/felixdittrich92/OnnxTR

Best,
Felix :)