[TF Lite] Re-train ssd_mobilenet_v1_quantized_coco

Resources

[1] How to quantify ssd_mobilenet_v1_coco model and toco to .tflite ? #18829

[2] SSD_mobilenet_v1/0.75_quantized_coco trained model is not detecting anything after porting on Android [Detect app] #21839

[3] COCO-trained models

 

方法论

一、开始训练

TF是个坑，但使用对的命令就可以了。

python object_detection/legacy/train.py --train_dir=training/ --pipeline_config_path=object_detection/ssd_mobilenet_v2_quantized_300x300_coco_2019_01_03/pipeline.config
python object_detection/legacy/train.py --train_dir=training/ --pipeline_config_path=object_detection/ssd_mobilenet_v1_coco_2018_01_28/pipeline.config
python object_detection/legacy/train.py --train_dir=training/ --pipeline_config_path=object_detection/ssd_mobilenet_v2_coco_2018_03_29/pipeline.config

ython object_detection/model_main.py --train_dir=training/ --pipeline_config_path=object_detection/ssd_mobilenet_v1_quantized_300x300_coco14_sync_2018_07_18/pipeline.config

 

If using rtx 2080, this code may be added for some tricky issues: Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR #34695

gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
    try:
        # Currently, memory growth needs to be the same across GPUs
        for gpu in gpus:
            tf.config.experimental.set_memory_growth(gpu, True)
        logical_gpus = tf.config.experimental.list_logical_devices('GPU')
        print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs")
    except RuntimeError as e:
        # Memory growth must be set before GPUs have been initialized
        print(e)

 

 

二、训练结果

/* implement */

 

 

三、模型转换

/* implement */

 

 

 

 

 

 

 

/* implement */