search

david8862 / keras-YOLOv3-model-set

end-to-end YOLOv4/v3/v2 object detection pipeline, implemented on tf.keras with different technologies
MIT License
639 stars 220 forks source link

saving model in tf format causes loading issues #230

Open mesolmaz opened 2 years ago

mesolmaz commented 2 years ago

I am trying to get away from saving the model as .h5 format, and saving it as saved_model format link. When I load the model using load_model call, I come across an error related to custom yolo_loss layer in yolo3. The model type I picked for training is yolo3_xception.

from tensorflow.keras.models import load_model

yolo_model_keras = "/workspaces/yolo/model" model = load_model(yolo_model_keras)

Here is the error code:

/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/layers/core.py:1045: UserWarning: yolo3.loss is not loaded, but a Lambda layer uses it. It may cause errors. , UserWarning) Traceback (most recent call last): File "main.py", line 4, in temp_model = load_model(yolo_model_keras) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/save.py", line 187, in load_model return saved_model_load.load(filepath, compile, options) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/saved_model/load.py", line 121, in load path, options=options, loader_cls=KerasObjectLoader) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/saved_model/load.py", line 633, in load_internal ckpt_options) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/saved_model/load.py", line 194, in init super(KerasObjectLoader, self).init(args, kwargs) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/saved_model/load.py", line 130, in init self._load_all() File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/saved_model/load.py", line 221, in _load_all self._finalize_objects() File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/saved_model/load.py", line 530, in _finalize_objects self._reconstruct_all_models() File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/saved_model/load.py", line 548, in _reconstruct_all_models self._reconstruct_model(model_id, model, layers) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/saving/saved_model/load.py", line 589, in _reconstruct_model config, created_layers={layer.name: layer for layer in layers}) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/functional.py", line 1214, in reconstruct_from_config process_node(layer, node_data) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/functional.py", line 1162, in process_node output_tensors = layer(input_tensors, kwargs) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py", line 926, in call input_list) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py", line 1117, in _functional_construction_call outputs = call_fn(cast_inputs, args, kwargs) File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/layers/core.py", line 903, in call result = self.function(inputs, kwargs) File "/workspaces/yolo/bin/yolo3/loss.py", line 294, in yolo3_loss grid, raw_pred, pred_xy, pred_wh = yolo3_decode(yolo_outputs[i], NameError: name 'yolo3_decode' is not defined

importing the following lines also doesn't help, the error changes to:

from yolo3.loss import yolo3_loss
from yolo3.postprocess import yolo3_decode

File "/workspaces/yolo/bin/yolo3/loss.py", line 295, in yolo3_loss anchors[anchor_mask[i]], num_classes, input_shape, scale_x_y=scale_x_y[i], calc_loss=True) TypeError: list indices must be integers or slices, not list

I believe this has to do with handling of custom layer during model build. Any idea how to save and load using this method?

david8862 commented 2 years ago

@mesolmaz not suggest to save the training checkpoint to save_model format. If hoping to use save_model for inference, you can dump out the checkpoint to inference model with TF save_model. see here