我嘗試將凍結(jié)的 SSD mobilenet v2 模型轉(zhuǎn)換為 TFLITE 格式以供 Android 使用。這是我所有的步驟：我使用來(lái)自模型動(dòng)物園的ssd_mobilenet_v2_coco_2018_03_29模型使用TF Object Detection API的 train.py 文件重新訓(xùn)練。（好的）使用同樣由 TF Object Detection API 提供的export_inference_graph.py將訓(xùn)練好的 model.ckpt 導(dǎo)出到凍結(jié)的模型文件。（好的）使用 GPU 測(cè)試 Python 中的凍結(jié)圖，并且僅允許使用 CPU。有用。（好的）缺點(diǎn)是，我嘗試使用以下代碼：import tensorflow as tftf.enable_eager_execution()saved_model_dir = 'inference_graph/saved_model/'converter = tf.contrib.lite.TFLiteConverter.from_saved_model(saved_model_dir,input_arrays=input_arrays,output_arrays=output_arrays,input_shapes={"image_tensor": [1, 832, 832, 3]})converter.post_training_quantize = True首先，我嘗試不向函數(shù)添加輸入形狀參數(shù)，但沒(méi)有奏效。從那時(shí)起我讀到你可以在那里寫(xiě)任何無(wú)關(guān)緊要的東西。直到這一行的輸出：INFO:tensorflow:Saver not created because there are no variables in the graph to restoreINFO:tensorflow:The specified SavedModel has no variables; no checkpoints were restored.INFO:tensorflow:The given SavedModel MetaGraphDef contains SignatureDefs with the following keys: {'serving_default'}INFO:tensorflow:input tensors info: INFO:tensorflow:Tensor's key in saved_model's tensor_map: inputsINFO:tensorflow: tensor name: image_tensor:0, shape: (-1, -1, -1, 3), type: DT_UINT8INFO:tensorflow:output tensors info: INFO:tensorflow:Tensor's key in saved_model's tensor_map: num_detectionsINFO:tensorflow: tensor name: num_detections:0, shape: (-1), type: DT_FLOATINFO:tensorflow:Tensor's key in saved_model's tensor_map: detection_boxesINFO:tensorflow: tensor name: detection_boxes:0, shape: (-1, 100, 4), type: DT_FLOATINFO:tensorflow:Tensor's key in saved_model's tensor_map: detection_scores我無(wú)法在此處復(fù)制控制臺(tái)輸出，因此它超過(guò)了 30000 個(gè)字符的限制，但您可以在這里看到它：https : //pastebin.com/UyT2x2Vk請(qǐng)?jiān)谶@一點(diǎn)上提供幫助，我該怎么做才能使其正常工作:(我的配置：Ubuntu 16.04，Tensorflow-GPU 1.12