INC new API tensorflow examples update

[lvliang-intel]

Signed-off-by: Lv, Liang1 [email protected]

Type of Change

feature
API changed

Description

detail description
JIRA ticket: ILITV-2456
INC New API TF examples(resnetv2_50,xception,inception_resnet_v2,resnet_v2)

Expected Behavior & Potential Risk

Tensorflow examples with partial build-in func or without build-in func, update to new API and Config class.

How has this PR been tested?

UT, Pre-CI test.

Dependency Change?

No.

[chensuyue]

1. delete config yaml.
2. unify data input with "--dataset_location=", we use the same dataset for calib and eval.
3. benchmark require "--mode=(accuracy/performance)", BenchmarkConfig is performance only.
   if mode==“accuracy”:
   eval_func()
   else:
   benchmark.fit(BenchmarkConfig(...), b_dataloader…)
4. Need to unify the why to handle ITEX gpu case. How to set gpu without yaml, with input param or ask user to add device=gpu in code?
5. Need to modify 'model_params_tensorflow.json', validation need to judge newAPI from the params list.
   "mode_name": {
   "model_src_dir": "xxx",
   "dataset_location": "xxx",
   "input_model": "xxx",
   "main_script": "xxx",
   "batch_size": 64
   },

[chensuyue]

extension test

newAPI: https://inteltf-jenk.sh.intel.com/job/intel-lpot-validation-top-mr-extension/3596/artifact/report.html
master oldAPI: https://inteltf-jenk.sh.intel.com/job/intel-lpot-validation-top-mr-extension/3598/artifact/report.html

issues

1. Tuning time consuming gap.
2. Please support benchmark for accuracy.

[chensuyue]

Extension test: https://inteltf-jenk.sh.intel.com/job/intel-lpot-validation-top-mr-extension/3630/artifact/report.html

[chensuyue]

extension test
pls check the performance gap.

[chensuyue]

extension test
minor issue need to fix, and pls rebuild the extension job with the failed model after bug fix.

[hshen14]

LGTM

[chensuyue]

still bas some performance gap, need to double check.

[chensuyue]

extension test