无法在 tensorflow 2 中导出 tflite_ssd_graph

Question

我使用 TensorFlow 对象检测训练模型 API。它工作正常我正在尝试为低端设备导出。我无法使用导出它。

python models/research/object_detection/export_tflite_ssd_graph.py 
--input_type image_tensor 
--pipeline_config_path output/training/001/pipeline.config 
--trained_checkpoint_prefix output/training/001/ckpt-40 
--output_directory exported-models/tf_lite_inference_model add_postprocessing_op=true

显示错误。

    export_tflite_ssd_graph_lib.export_tflite_graph(
  File "C:\Users\inspiron\anaconda3\lib\site-packages\object_detection\export_tflite_ssd_graph_lib.py", line 241, in export_tflite_graph
    image = tf.placeholder(
  File "C:\Users\inspiron\anaconda3\lib\site-packages\tensorflow\python\ops\array_ops.py", line 3176, in placeholder
    raise RuntimeError("tf.placeholder() is not compatible with "

Answer 1

安装tf-nightly即可解决

python -m pip install tf-nightly

如需更多参考，请查看 Tensorflow Model Repo 上的 Issue。

无法在 tensorflow 2 中导出 tflite_ssd_graph

Unable to export tflite_ssd_graph in tensorflow 2

python

tensorflow

object-detection-api