模型量化后初始化失败

[float123]

1. 使用环境（environment）

• centos，cuda10.2，cudnn 8.2.0，tensorrt7.2.1

2. Github版本

• branch：master
• commit(optional):

3. 详细描述bug 情况 (Describe the bug) TNN版本的crnn可以正常运行，尝试将其量化 量化命令： /raid/TNN/platforms/linux/build ./quantization_cmd -p /raid/tnn_ocr/model/CRNN_bilstm.tnnproto -m /raid/tnn_ocr/model/CRNN_bilstm.tnnmodel -i /raid/tnn_ocr/crnn_img -b 0,0,0 -s 0.0039216,0.039216,0.0039216 加载量化后的模型进行初始化的时候会报错。

4. 日志（Log） E/tnn: Init [File /raid/shuike/TNN/TNN/source/tnn/layer/base_layer.cc][Line 57] InferOutputShape failed E/tnn: InitLayers [File /raid/shuike/TNN/TNN/source/tnn/network/tensorrt/tensorrt_network.cc][Line 454] Error Init layer 82 (err: 16384 or 0x4000) I/tnn: Init [File /raid/shuike/TNN/TNN/source/tnn/core/instance.cc][Line 77] Init network failed. Try to re-init it with const folder, and if succeed all of error info above can be ignored. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: 219_input_quant_: at least 4 dimensions are required for input. ERROR: Layer 219_input_quant_ failed validation ERROR: Network validation failed.

5. 截图（Screenshots） 量化后的模型和原来的模型似乎发生了改变

[float123]

量化后的模型 model.quantized.zip