* graph.input can be either ValueInfoProto or string

* pylint

* Change the way to import ONNX model.

* Decide ONNX version and change the processing

* Check whether onnx has the attribute '__version__'

* Correct the use of `concatenate` operator

* Optimize

* fix for composed symbol

* fix

* clean up

* fix exception type

* fix recently released layers

* fix fc layers with partial infer_shape

* add coreml

* fix bool

* fix flatten in fullyconnected

* fix duplicate flatten

* fix syntax

* add tutorial

* fix mxnet flatten, fix tutorials

* fix flatten issue

* fix lint

* resnet example merged to imagenet

* resnet example merged to imagenet

* merge with master

* merge with master

* cuda imagenet benchmakr added

* benchmark syntax fixed

* [TEST] include coreml tools in docker

* upgrade tvm

* [BENCH] Add Benchmark for Rasp

* [BENCH] Add arg opt-level

* [BENCH] Add model choices

* [BENCH] Improve

* [TUTORIAL] Deploy the Pretrained Model on Raspberry Pi

* [TUTORIAL] Improve

* [TUTORIAL] Improve

* [TUTORIAL] Improve

* [TUTORIAL] Improve

* add json spec intro

* fix error

* fix name attrs

* rename and move to top

* address boolean values

* update title

* update

* back to color

* [TUTORIAL] ImageNet Inference on Raspberry Pi

* Update tvm

* [PASS] FoldScaleAxis

* Move FoldAxis to O3

* Set unroll to 0 when ready

* add transform

* fix

* update doc

* Update tvm

* [TEST] Xavie initialization for benchmarks

* remove additional line

* use resnet v2

* fix

* add onnx

* fix