* cpu format shape
* convert model
* python api
* speedup dataloader
* minor
* add cpp demo
* add visualize
* fix resize param order
* export pipeline.json
* fix three crop
* read SampleFrames from model_cfg
* minor
* lint
* move to a func
* speed up format shape cpu
* use input mat device
* fix comments
* fix comments
* update docs/benchmark
* docs/supported-codebases
* update tests/data
* fix lint
* fix lint
