Tensoflow目标检测实战（4）训练模型转换至tflite并部署

官方参考笔记

..../models/research/object_detection/colab_tutorials/convert_odt_model_to_TFLite.ipynb

1.  下载预训练模型或自己数据训练得到的模型

wget http://download.tensorflow.org/models/object_detection/tf2/20200711/ssd_mobilenet_v2_fpnlite_640x640_coco17_tpu-8.tar.gz

以mobilenetV模型为例，解压后文件包括 ：

根目录

checkpoint目录（模型主要文件）

 

 2. 输出pb模型文件，inference graph

python object_detection/export_tflite_graph_tf2.py \
    --trained_checkpoint_dir ssd_mobilenet_v2_fpnlite_640x640_coco17_tpu-8/checkpoint \
    --output_directory ssd_mobilenet_v2_fpnlite_640x640_coco17_tpu-8/tflite \
    --pipeline_config_path ssd_mobilenet_v2_fpnlite_640x640_coco17_tpu-8/pipeline.config

运行结果

2022-09-26 00:33:04.424700: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:04.468000: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:04.468301: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:04.479873: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2022-09-26 00:33:04.480393: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:04.480556: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:04.480691: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:05.261160: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:05.261287: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:05.261350: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:05.261616: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 9858 MB memory:  -> device: 0, name: NVIDIA GeForce RTX 3060, pci bus id: 0000:01:00.0, compute capability: 8.6
2022-09-26 00:33:08.810228: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:08.810660: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:08.810739: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:08.810834: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:08.810890: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:08.810943: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 9858 MB memory:  -> device: 0, name: NVIDIA GeForce RTX 3060, pci bus id: 0000:01:00.0, compute capability: 8.6
2022-09-26 00:33:09.697849: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:09.697982: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:09.698051: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:09.698146: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:09.698216: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2022-09-26 00:33:09.698272: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 9858 MB memory:  -> device: 0, name: NVIDIA GeForce RTX 3060, pci bus id: 0000:01:00.0, compute capability: 8.6
WARNING:tensorflow:Skipping full serialization of Keras layer <object_detection.meta_architectures.ssd_meta_arch.SSDMetaArch object at 0x7f6f70070eb8>, because it is not built.
W0926 00:33:09.972333 140118163120320 save_impl.py:72] Skipping full serialization of Keras layer <object_detection.meta_architectures.ssd_meta_arch.SSDMetaArch object at 0x7f6f70070eb8>, because it is not built.
WARNING:tensorflow:Skipping full serialization of Keras layer <keras.layers.convolutional.SeparableConv2D object at 0x7f6f5c37d978>, because it is not built.
W0926 00:33:10.072510 140118163120320 save_impl.py:72] Skipping full serialization of Keras layer <keras.layers.convolutional.SeparableConv2D object at 0x7f6f5c37d978>, because it is not built.
WARNING:tensorflow:Skipping full serialization of Keras layer <object_detection.core.freezable_batch_