search

CasiaFan / SSD_EfficientNet

SSD using TensorFlow object detection API with EfficientNet backbone
63 stars 10 forks source link

Error while running train.py #16

Open victor-yudin opened 4 years ago

victor-yudin commented 4 years ago

After performing the steps from instructions here I try to run training ssd efficientnet fpn on coco ms dataset:

%run legacy/train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_efficientdet.config

Error occured:

    200     })
    201 
--> 202   if feature_extractor_config.HasField('num_layers'):
    203     kwargs.update({'num_layers': feature_extractor_config.num_layers})
    204 

ValueError: Protocol message SsdFeatureExtractor has no field num_layers.

I Also tried to comment this condition in model_builder.py, but got another error (Attribute error).

If to switch the protos back, the "network_version" ValueError occurs.