Created by: bingyanghuang

Paddle develop branch INT8 performance regression problem has not been solved yet. This regression is caused by merged PR #17965 , commit number is 84bb45c0 by @jczaja . Regression data is shown in the following table: Testing command line is ./paddle/fluid/inference/tests/api/test_analyzer_int8_image_classification --infer_model=third_party/inference_demo/int8v2/resnet50/model --infer_data=/~/.cache/paddle/dataset/int8/download/int8_full_val.bin --batch_size=1 --paddle_num_threads=1 You can change --infer_model with different model name googlenet, mobilenetv1, mobilenetv2, resnet101, resnet50, vgg16, vgg19 for different benchmarks, refer to int8v2_readme.