First to belle is OpenVINO written document is extremely poor, almost can not find anything you want, perhaps this is the essence of the open source project, it's up to see the code... So think about it first, when I am with DL workbench, can convert mobilenet - SSD model and test accuracy, so this part must be take in the openvino installation directory, flip openvino accuracy_checker directory, saw a suspicious file
Look at the file content
Models:
- name: SSD - mobilenet - cf
Launchers:
- framework: caffe
Model: public/mobilenet - SSD/mobilenet - SSD. Prototxt
Weights: public/mobilenet - SSD/mobilenet - SSD. Caffemodel
Adapter: SSD
Datasets:
- name: VOC2012
Preprocessing:
- type: resize
Size: 300
- type: normalization
Mean: 127.5
STD: 127.5
Postprocessing:
- type: resize_prediction_boxes
- name: SSD - mobilenet
Launchers:
- framework: DLSDK
Tags:
- FP32
Model: public/mobilenet - SSD/FP32/mobilenet - SSD. XML
Weights: public/mobilenet - SSD/FP32/mobilenet - SSD. Bin
Adapter: SSD
- framework: DLSDK
Tags:
- FP16
Model: public/mobilenet - SSD/FP16/mobilenet - SSD. XML
Weights: public/mobilenet - SSD/FP16/mobilenet - SSD. Bin
Adapter: SSD
Datasets:
- name: VOC2012
Preprocessing:
- type: resize
Size: 300
Postprocessing:
- type: resize_prediction_boxes
Global_definitions:../dataset_definitions yml
Launchers section, there is a model of the path and file name; Datasets part, writing the data set is VOC2012, under part of the description is not very full, but here is a strange thing: global_definitions:../dataset_definitions yml
Looks a bit like C include, see the file
In the file content to find VOC2012 keyword
- name: VOC2012
Annotation_conversion:
The converter: voc_detection
Annotations_dir: VOCdevkit VOC2012/Annotations
Images_dir: VOCdevkit VOC2012/JPEGImages
Imageset_file: VOCdevkit/VOC2012/ImageSets/Main/val. TXT
Data_source: VOCdevkit VOC2012/JPEGImages
Annotation: voc12 pickle
Dataset_meta: voc12 json
Postprocessing:
- type: resize_prediction_boxes
The metrics:
- type: map
11 point integral:
Ignore_difficult: True
Presenter: print_scalar
This has been all about, create a mobilenet - ssd_ac. Yml file, use Ctrl CV solution, all the things I want to copy and paste together
Models:
- name: SSD - mobilenet
Launchers:
- framework: DLSDK
Tags:
- FP32
Model: XML mobilenet - SSD.
Weights: mobilenet - SSD. Bin
Adapter: SSD
Datasets:
- name: VOC2012
Preprocessing:
- type: resize
Size: 300
Postprocessing:
- type: resize_prediction_boxes
Annotation_conversion:
The converter: voc_detection
Annotations_dir: VOCdevkit VOC2012/Annotations
Images_dir: VOCdevkit VOC2012/JPEGImages
Imageset_file: VOCdevkit/VOC2012/ImageSets/Main/val. TXT
Data_source: VOCdevkit VOC2012/JPEGImages
Annotation: voc12 pickle
Dataset_meta: voc12 json
Postprocessing:
- type: resize_prediction_boxes
The metrics:
- type: map
11 point integral:
Ignore_difficult: True
Presenter: print_scalar
Then convert mobilenet - IR model of SSD
C: \ Program Files \ (x86) IntelSWTools \ openvino \ deployment_tools \ model_optimizer & gt; Python mo_caffe. Py -- -- input_model c: \ work \ cvt_model \ mobilenet - SSD caffemodel - o c: \ work \ mobilenet - 127.5 - mean_values ssd_ac - scale [127.5, 127.5, 127.5]
Again from the Internet under a VoC model VOCtrainval_11 - May - 2012. The tar
The several files together
Run it and see
The mAP accuracy is 79.84% basic can prove this config file is written by the way:)
To summarize:
In C: \ Program Files \ IntelSWTools \ openvino_2020 (x86) 3.194 \ deployment_tools \ open_model_zoo \ tools \ accuracy_checker \ configs here are many pre-written open source model accuracy check configuration Files, writing their own model checking configuration can be used to refer to it:)