experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Lite] num_ranks option brings to user through cmd line parameter

Browse Source 
num_ranks optional parameters provides to user to calculate accuracy ranks on out file.
This commit is contained in:
Dayananda-V 2019-07-04 19:32:00 +05:30
parent ad6e2c74ce
commit 668bd2a76f
2 changed files with 8 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/lite/tools/accuracy/ilsvrc/README.md
View File

@ -56,6 +56,10 @@ and the following optional parameters:
Optionally, the computed accuracies can be output to a file as a
string-serialized instance of tflite::evaluation::TopkAccuracyEvalMetrics.
* `num_ranks`: `int` (default=10) \
The number of top-K accuracies to return. For example, if num_ranks=5,
top-1 to top-5 accuracy fractions are returned.
The following optional parameters can be used to modify the inference runtime:
* `num_interpreter_threads`: `int` (default=1) \

4
tensorflow/lite/tools/accuracy/ilsvrc/imagenet_model_evaluator.cc
View File

@ -49,6 +49,7 @@ constexpr char kInterpreterThreadsFlag[] = "num_interpreter_threads";
constexpr char kDelegateFlag[] = "delegate";
constexpr char kNnapiDelegate[] = "nnapi";
constexpr char kGpuDelegate[] = "gpu";
constexpr char kNumRanksFlag[] = "num_ranks";
template <typename T>
std::vector<T> GetFirstN(const std::vector<T>& v, int n) {
@ -144,6 +145,9 @@ class CompositeObserver : public ImagenetModelEvaluator::Observer {
tflite::Flag::CreateFlag(kDelegateFlag, &params.delegate,
"Delegate to use for inference, if available. "
"Must be one of {'nnapi', 'gpu'}"),
tflite::Flag::CreateFlag(kNumRanksFlag, &params.num_ranks,
"Generates the top-1 to top-k accuracy values"
"where k = num_ranks. Default: 10"),
};
tflite::Flags::Parse(&argc, const_cast<const char**>(argv), flag_list);