paddleservering部署自训练模型失败，如何修改

[charm8686]

1. PaddleClas release/2.3和PaddleClas release/2.4都试过。PaddlePaddle 2.2.0和PaddlePaddle 2.2.2和PaddlePaddle 2.3.1都试过
2. PaddleServing==0.7.0和0.9.0都试过
3. 训练环境信息： a. ubuntu18 paddle推荐docker b. Python3.6 c. CUDA11.2/cuDNN 8.0.0等
4. 使用百度提供ResNet50_vd预训练模型训练自定义数据集模型，在使用paddle_serving_client.convert转换模型后，按照https://github.com/PaddlePaddle/PaddleClas/tree/release/2.4/deploy/paddleserving文档部署后一直报错。
5. 报错日志： ERROR 2022-07-11 22:18:18,717 [operator.py:979] (data_id=0 log_id=0) [imagenet|0] Failed to process(batch: [0]): 'x' Traceback (most recent call last): File "/home/miniconda3/envs/clas/lib/python3.6/site-packages/paddle_serving_server/pipeline/operator.py", line 966, in _run_process feed_batch, typical_logid) File "/home/miniconda3/envs/clas/lib/python3.6/site-packages/paddle_serving_server/pipeline/operator.py", line 577, in process log_id=typical_logid) File "/home/miniconda3/envs/clas/lib/python3.6/site-packages/paddle_serving_app/local_predict.py", line 281, in predict if isinstance(feed[name], list): KeyError: 'x' ERROR 2022-07-11 22:18:18,718 [operator.py:1024] (log_id=0) imagenet failed to predict. ERROR 2022-07-11 22:18:18,720 [dag.py:410] (data_id=0 log_id=0) Failed to predict: (log_id=0) imagenet failed to predict.
6. 打印feed为inputs，但是模型输入为'x'，即使改了serving_server_conf.prototxt的name也无用

[RainFrost1]

已经更新，具体请查看最新pr

[HydrogenSulfate]

可以尝试将classification_web_service.py的{"inputs": input_imgs}改为{"x": input_imgs}，保证你导出的serving_server_conf.prototxt文本里feed_var的name与上述代码中的一致