Problem in load_model function

[shriyavanvari]

I get the following error on running:

Traceback (most recent call last):
File "main.py", line 30, in
main(arg_parser.parse_args())
File "main.py", line 13, in main
yolo = YOLO(config)
File "/Users/shriyavanvari/Documents/SJSU/Semester 1/CMPE 257/CARLA/Code/traffic-light-detection-module/yolo.py", line 113, in init
self.feature_extractor = TinyYoloFeature(self.image_h, config).feature_extractor
File "/Users/shriyavanvari/Documents/SJSU/Semester 1/CMPE 257/CARLA/Code/traffic-light-detection-module/yolo.py", line 56, in init
pretrained = load_model('checkpoints/' + config['model']['saved_model_name'], custom_objects={'custom_loss': dummy_loss, 'tf': tf})
File "/Users/shriyavanvari/anaconda3/lib/python3.7/site-packages/keras/engine/saving.py", line 492, in load_wrapper
return load_function(*args, **kwargs)
File "/Users/shriyavanvari/anaconda3/lib/python3.7/site-packages/keras/engine/saving.py", line 584, in load_model
model = _deserialize_model(h5dict, custom_objects, compile)
File "/Users/shriyavanvari/anaconda3/lib/python3.7/site-packages/keras/engine/saving.py", line 270, in _deserialize_model
model_config = h5dict['model_config']
File "/Users/shriyavanvari/anaconda3/lib/python3.7/site-packages/keras/utils/io_utils.py", line 318, in getitem
raise ValueError('Cannot create group in read-only mode.')
ValueError: Cannot create group in read-only mode.
Can you please help out?

[filipbasara0]

Hi,

Can you please specify which Tensorflow and Keras versions you are using?

[shriyavanvari]

Keras 2.3.1
Tensorflow 1.14.0

[filipbasara0]

Can you try using Keras 2.2.4 and Tensorflow 1.12.0? I tested it out with those versions and it seemed to be working fine.

[spadesM]

this error happened to me because there is no pretrained model.
"pretrained = load_model('checkpoints\' + config['model']['saved_model_name'], custom_objects={'custom_loss': dummy_loss, 'tf': tf})"

you can find from description.

[Kangsan-Jeon]

When I ran 'detect_test_images.py' with Windows10, keras(2.2.4) and tensorflow(1.15.0), error occurred 'load_model' operation. The error was 'ValueError: Cannot create group in read only mode.'. In my case, I made 'checkpoints' directory and put your 'traffic-light-detection.h5' file in 'checkpoints' folder. How can I solve this problem?

I also modified json file like below.

{
"model" :
{
"image_h": 416,
"image_w": 416,
"anchors": [0.24,0.79, 0.80,2.12],
"max_obj": 5,
"grid_h": 13,
"grid_w": 13,
"num_classes": 2,
"classes": ["go", "stop"],
"obj_thresh": 0.20,
"nms_thresh": 0.01,
"saved_model_name": "traffic-light-detection.h5",
"obj_scale": 85,
"no_obj_scale": 75,
"coord_scale": 70,
"class_scale": 70
},

"plot_model":           true,

"train": {
    "enabled":              false,
    "annot_file_name":      "carla_all.csv",
    "batch_size":           8,
    "learning_rate":        1e-4,
    "nb_epochs":            500,
    "warmup_batches":       0,
    "debug":                true
}

}

[masskro0]

I have the same problem as @Kangsan-Jeon .

[masskro0]

Ok I could fix this issue. You need to access the model attribute of the YOLO class, e.g.
model = YOLO(config)
and then you can call
yolo.model.load_weights("checkpoints\\traffic-light-detection.h5")
That is how you load the pretrained model. But you have to make several changes to the code.
In yolo.py in the TinyYoloFeature class, i removed the following lines:
pretrained = load_model('checkpoints\\' + config['model']['saved_model_name'], custom_objects={'custom_loss': dummy_loss, 'tf': tf})
pretrained = pretrained.get_layer('model_1')
idx = 0
for layer in self.feature_extractor.layers:
print(layer.name)
layer.set_weights(pretrained.get_layer(index=idx).get_weights())
idx += 1

as well as in the YOLO class the following lines:
pretrained = load_model('checkpoints\\' + config['model']['saved_model_name'], custom_objects={'custom_loss': self.custom_loss, 'tf': tf})
self.model.get_layer('DetectionLayer').set_weights(pretrained.get_layer('DetectionLayer').get_weights())

Now none of these classes loads the model and you can load the pretrained model as I described above. Hope this helps.

[aeermis]

hello, I get the following error, could you please help me about it?

This function in the predict.py file;

1- def get_model_from_file(config):
2- path = os.path.join(BASE_DIR, 'checkpoints', config['model']['saved_model_name'])
3- model = load_model(path, custom_objects={'custom_loss': dummy_loss})
4- return model

I get the error:

SavedModel file does not exist at: C:\Users\90531\Source\Repos\traffic-light-detection-module\checkpoints\model.09-2.01.h5/{saved_model.pbtxt|saved_model.pb}