1 # Deep Learning accuracy validation framework
5 You may test your installation and get familiar with accuracy checker by running [sample][sample-readme].
7 Once you installed accuracy checker you can evaluate your configurations with:
10 python3 accuracy_check.py -c path/to/configuration_file -m /path/to/models -s /path/to/source/data -a /path/to/annotation
13 All relative paths in config files will be prefixed with values specified in command line:
15 - `-c, --config` path to configuration file.
16 - `-m, --models` specifies directory in which models and weights declared in config file will be searched.
17 - `-s, --source` specifies directory in which input images will be searched.
18 - `-a, --annotations` specifies directory in which annotation and meta files will be searched.
20 You may refer to `-h, --help` to full list of command line options. Some optional arguments are:
22 - `-e, --extensions` directory with InferenceEngine extensions.
23 - `-b, --bitstreams` directory with bitstream (for Inference Engine with fpga plugin).
24 - `-C, '--converted_models` directory to store Model Optimizer converted models (used for DLSDK launcher only).
25 - `-tf, --target_framework` framework for infer.
26 - `-td, --target_devices` devices for infer. You can specify several devices using space as a delimiter.
30 There is config file which declares validation process.
31 Every validated model has to have its entry in `models` list
32 with distinct `name` and other properties described below.
34 There is also definitions file, which declares global options shared across all models.
35 Config file has priority over definitions file.
44 model: public/alexnet/caffe/bvlc_alexnet.prototxt
45 weights: public/alexnet/caffe/bvlc_alexnet.caffemodel
46 adapter: classification
54 Launcher is a description of how your model should be executed.
55 Each launcher configuration starts with setting `framework` name. Currently *caffe* and *dlsdk* supported. Launcher description can have differences.
59 - [how to configure Caffe launcher][caffe-launcher-configuration].
60 - [how to configure DLSDK launcher][dlsdk-launcher-configuration].
64 Dataset entry describes data on which model should be evaluated,
65 all required preprocessing and postprocessing/filtering steps,
66 and metrics that will be used for evaluation.
68 If your dataset data is a well-known competition problem (COCO, Pascal VOC, ...) and/or can be potentially reused for other models
69 it is reasonable to declare it in some global configuration file (*definition* file). This way in your local configuration file you can provide only
70 `name` and all required steps will be picked from global one. To pass path to this global configuration use `--definition` argument of CLI.
72 Each dataset must have:
74 - `name` - unique identifier of your model/topology.
75 - `data_source`: path to directory where input data is stored.
76 - `metrics`: list of metrics that should be computed.
79 - `preprocessing`: list of preprocessing steps applied to input data. If you want calculated metrics to match reported, you must reproduce preprocessing from canonical paper of your topology or ask topology author about required steps if it is ICV topology.
80 - `postprocessing`: list of postprocessing steps.
81 - `reader`: approach for data reading. You can specify: `opencv_imread` or `pillow_imread` for reading images and `opencv_capture` for reading frames from video. Default reader is `opencv_imread`.
83 Also it must contain data related to annotation.
84 You can convert annotation inplace using:
85 - `annotation_conversion`: parameters for annotation conversion
88 or use existing annotation file and dataset meta:
89 - `annotation` - path to annotation file, you must **convert annotation to representation of dataset problem first**, you may choose one of the converters from *annotation-converters* if there is already converter for your dataset or write your own.
90 - `dataset_meta`: path to metadata file (generated by converter).
91 More detailed information about annotation conversion you can find [here][converters]
93 example of dataset definition:
97 annotation: annotation.pickle
98 data_source: images_folder
105 - type: normalization
116 ### Preprocessing, Metrics, Postprocessing
118 Each entry of preprocessing, metrics, postprocessing must have `type` field,
119 other options are specific to type. If you do not provide any other option, then it
120 will be picked from *definitions* file.
122 You can find useful following instructions:
124 - [how to use preprocessings][preprocessors].
125 - [how to use postprocessings][postprocessors].
126 - [how to use metrics][metrics].
128 You may optionally provide `reference` field for metric, if you want calculated metric
129 tested against specific value (i.e. reported in canonical paper).
131 Some metrics support providing vector results ( e. g. mAP is able to return average precision for each detection class). You can change view mode for metric results using `presenter` (e.g. `print_vector`, `print_scalar`).
143 ### Testing new models
145 Typical workflow for testing new model include:
147 1. Convert annotation of your dataset. Use one of the converters from annotation-converters, or write your own if there is no converter for your dataset. You can find detailed instruction how to use converters [here][converters].
150 python3 convert_annotation.py converter --converter_specific_parameter --output_dir data/annotations
153 1. Choose one of *adapters* or write your own. Adapter converts raw output produced by framework to high level problem specific representation (e.g. *ClassificationPrediction*, *DetectionPrediction*, etc).
154 1. Reproduce preprocessing, metrics and postprocessing from canonical paper.
155 1. Create entry in config file and execute.
157 [sample-readme]: ./tools/accuracy_checker/sample/README.md
158 [preprocessors]: ./tools/accuracy_checker/accuracy_checker/preprocessor/README.md
159 [postprocessors]: ./tools/accuracy_checker/accuracy_checker/postprocessor/README.md
160 [metrics]: ./tools/accuracy_checker/accuracy_checker/metrics/README.md
161 [converters]: ./tools/accuracy_checker/accuracy_checker/annotation_converters/README.md
162 [caffe-launcher-configuration]: ./tools/accuracy_checker/accuracy_checker/launcher/caffe_launcher_readme.md
163 [dlsdk-launcher-configuration]: ./tools/accuracy_checker/accuracy_checker/launcher/dlsdk_launcher_readme.md