Releases: JDAI-CV/DNNLibrary
Releases · JDAI-CV/DNNLibrary
v0.9.0: Merge pull request #71 from JDAI-CV/fix_exception_in_ort
Changes:
- e17f11e Merge pull request #71 from JDAI-CV/fix_exception_in_ort
- 4efa112 Merge branch 'master' into fix_exception_in_ort
- a3bfb68 catch exceptions in GetSupportedNodes so that ort nnapi ep will not crash here
- bac3cce Merge pull request #70 from JDAI-CV/add_ops
- e47f9df Add three ops: sub, exp and abs, and update some code
- af84537 Merge pull request #69 from JDAI-CV/ort_ci
- c2dbf71 Revert "Intentionally use a wrong floor op impl to test the ci"
- e59a642 Intentionally use a wrong floor op impl to test the ci
- ac0e086 use daquexian android ort branch
- 656180e Fix ort source dir
See More
- dd96d39 Fix ci yml
- 49ead95 Add ONNX Runtime CI
- ad11c32 Merge pull request #67 from JDAI-CV/update_script_add_new_ops
- 2980add Fix nnapi Bad data error in api < 29 conv_2d
- f64b511 Support nnapi inputs that are only available in some api levels, update conv_2d
- 5aee023 Remove unused file
- 79ac7cb delete copy constructor and copy assignment op of Model according to the rule of three
- 0ff3719 Use strong typed FuseCode all the places
- f779d6c Refactor code about fuse code
- 36c15fb Split fbs param into input param and output param, for better fitting NNAPI
- 5bd1c19 Remove some throw in mbimpl.cpp
- b188392 Change some methods of daq reader to return expected
- a63722c clang-format ex_model_builder.cpp
- 4cd010e Update api names in OnnxConverter
- 512b539 Make AddLayer_* in model builder return expected object instead of throw exceptions on error
- 594ee33 Update the api names in ex_model_builder.cpp
- 3500f1a Add blank lines between generate methods in MBImpl.cpp
- d8b8081 Merge branch 'master' into update_script_add_new_ops
- 4fad420 Rename AddLayer* to AddLayer_* and WriteDaqLayer_*
- eb620ee Rename the misleading attr name in ops.yml
- 3c82a70 introduce expected.hpp
- 66f9a2b Fix the unused
buffer
bug - 1c5cee0 Relax the cmake version for android ndk
- a5d5627 Relax the cmake version for android ndk
- e97b9db Make AddLayer*Impl private
- fa28ade ModelBuilder AddLayer* return void now
- 96e4721 Support prelu before android 29
- 19f7f49 Rename Add* to AddLayer* in ModelBuilder
- 8b547c2 Rename AddTensorFromBuffer to ...Persistent..., AddTensorFromBuffer now copies the data
- 37e92bd Add version field and refuse old models
- c7976bb Fix wrong kernel shape of global pooling
- 87466ec make ci log less verbose
- 0895bb4 'converter_simple' should not exist, all related logic should be implemented in Convert()
- f46f008 fix wrong check and re-format onnxconverter.cpp
- 45b0c1b Merge branch 'master' into update_script_add_new_ops
- 4e6078c Make most boilderplate code auto-generated, change many layer names, old daq models will not work any more
- 47caec4 Fix the wrong axis order, try to move all logic into onnxconverter Convert()
- e0dd8ab Add compile_fbs() in generate_code.py, change shaper of min/maximum to eltwise
- e637232 Add the missing auto-generated code
- f498eac Remove "=0" since default value is 0
- 4041d0b Update generate_code.py to generate daq.fbs
- 77520e1 Ignore .clangd and venv/
- b72182e Add some new ops in ops.yml
- 9fdc96f Add a little comment in ops.yml
- dbb8e3f Improve some header include path
- 8370c6f Remove out-dated benchmark in README
- 3c77c30 Merge pull request #65 from JDAI-CV/update_readme
- 0d9e665 Update benchmark in README
- 6612d1e Install cmake 3.14.4 in CI
- 647d4c3 Refuse to build DNNLibrary with cmake 3.15
- 2d48e27 Update PRs welcome badge style
- 2d4b0f0 check api in isnodesupported()
- 52abfa8 Merge pull request #63 from JDAI-CV/update
- c969332 Support sigmoid
- e17f618 Check Android SDK in runtime instead of compile time
- 156eab7 Generate daqreader layer_type_to_str in generate_code.py
- 051b1be The missing rename in generate_code.py
- abcc3b5 not .. in -> .. not in
- 70813c6 format generate_code.py
- 7e70872 Merge pull request #62 from JDAI-CV/update
- 589611c Move template function
- 5f76678 Remove unused function declaration in ModelBuilder.h
- d093f6b Add missing code in DaqReader
- 2719734 Move op implementation of OnnxConverterImpl into OnnxConverterImpl.cpp
- b5dd437 Support two simple op: tanh and floor
- 77fa973 Move op implementation to ModelBuilderImpl.cpp
- b7073db Add comment for cmake 3.15 workaround
- c3cd873 Merge pull request #61 from JDAI-CV/update
- 29a6a99 have no idea why cmake_cxx_standard doesn't work suddenly
- b80ab3c getDevices -> GetDevices
- 812bbe0 Fix std::nullopt
- 4923e9f Add getDevices()
- cf9c566 Remove redundant shaper functions
This list of changes was auto generated.
v0.7.3: Merge pull request #60 from JDAI-CV/update_wrapper
Changes:
- 848b29f Merge pull request #60 from JDAI-CV/update_wrapper
- 4168162 Update nnapi wrapper from tensorflow
- ab22710 Merge pull request #59 from JDAI-CV/check_conv_weight
- a724168 Use shape inference to check concat, softmax and global pooling
- ecc612d Add more check for pooling, batchnorm and identity
- c28e516 getsupportednodes.cpp building -- fix cmake
- 73e4e89 Fix the shape of one-element tensor
- eff13f6 Add rank check in convert func of onnx converter
- 4a4ded8 Merge pull request #58 from JDAI-CV/check_conv_weight
- 85cad29 Add getsupportednodes tool, check for single reshape/dropout
See more
- 79edcc3 Add softmax axis check
- 8501d9a fix depth mulitiplier of dw conv
- 2f08809 conv2d and pooling2d
- b1a25f6 make strkeymap has const, check prelu slope
- 51f2779 check whether onnx convolution weight exists in initializer
- 65b14c9 check api in runtime and make AllowFp16 accessible in api 27
- a804ce1 remove src folder
- efa45e7 make log less verbose
- b5739d5 Update dnn_benchmark to support onnx
- b9584fe Support non-rank-4 input in daqreader
- 1ec5eaa Add the missing throw in dnn_retrieve_result.cpp
- 90cb0f8 Merge pull request #57 from JDAI-CV/support_onnx_model_output
- a75548d Considering identity for output
- 6bbf563 support onnx model output
- 99baa7b Merge pull request #56 from JDAI-CV/3rd_party
- 73f7526 Move flatbuffers to submodule
This list of changes was auto generated.
v0.7.2: Merge pull request #55 from JDAI-CV/support_onnx_input
Changes:
- cf20ec9 Merge pull request #55 from JDAI-CV/support_onnx_input
- 550c68f Fix gemm weight conversion
- 7c1d1b9 Fix eltwise shaper
- 38dd3bd Support onnx input (which can be read from both output and initializer)
- 8cab557 Fix warnings
- b33f375 Merge pull request #54 from JDAI-CV/getsupportednodes
- 869d560 Add GetSupportedNodes()
- 9120998 Add IsNodeSupported()
This list of changes was auto generated.
v0.7.1
Changes:
- 2a705eb Apply clang-format
- 8ded267 Add missing headers
- 3ac1034 Remove android header in onnx2daq
- 1f10295 Update the check of auto_pad
- 4d918b9 Fix bug that input width and height are reversed
- 276f2dd Use dnn::optional instead of std::optional and nonstd::optional
- 59646a6 Update path in generate_code.py
- b5d6707 Fix bug in dnn_retrieve_result.cpp
- 7f5d2f7 Use protobuf-lite in onnx2daq
- 013c6a2 Use protobuf-lite instead of protobuf
See more
- 4fe4b42 Add DNN_CMAKE_INSTALL for disabling cmake install in onnxruntime
- 52ad7de Add guard to not add_subdirectory when the target exists
- 52da80a Add missing DNN_READ_ONNX guard
- 1014a45 Merge pull request #53 from JDAI-CV/export
- 7f794cf Fix alias and destination
- 5ab120f Add alias the same with exported target name
- 2f96e4c Add destination for installing and fix jni
- a6fd498 Add dnn:: namespace
- 6ea82ea Update Config.cmake.in and so on
- 34b5fa2 Reorg the directory structure for exporting, also remove helper.h in
- 51b19b2 Auto set python package version from git tag
This list of changes was auto generated.
v0.7.0
Changes:
- bf69edf Auto generate versionCode in android build.gradle
- 9b428af Avoid name conflict in glog.cmake. BUILD_TESTING is conflicted with eigen in onnxruntime
- 4f12fe1 Merge pull request #52 from JDAI-CV/ci
- 3ca03f7 Update README badge
- a1a295d Skip windows ci
- de62356 Add all-27 configuration
- e1522d5 Fix wrong path of libdaq-jni.so
- 7e86fd8 Fix undefined AllowFp16 in Android 27
- f2fd69a Fix wrong cmake args, delete travis and appveyor ci
- 9a057f3 Fix a bug found by CI: resnetv2 fails
See more
- 748ddd8 Install onnx in ci
- faaf1d2 Disable werror for binaries when bundled protobuf is used
- 2cf973c Fix unused parameter in dnn_benchmark.cpp
- 3d13eab Disable treat warnings as errors for dnnlibrary when reading onnx
- e4ef696 Not build protoc on Android
- c13a6e7 Fix protoc path
- 5182c6d Fix protoc path
- fd53321 Download protoc in ci
- 202aefb fix msvc flag
- db6c783 x86_64 emulator
- e835863 Set github connection
- cdede50 Set android api for dnnlibrary build
- 5376b8a Edit ci files to fit dnnlibrary
- bd9bfd1 rename dabnn to dnnlibrary, onnx2bnn to onnx2daq
- ce1a521 azure pipeline ci init -- move files from dabnn
- a645f98 Merge pull request #47 from JDAI-CV/python_package
- 26af0fb Remove redundant find protobuf
- 3d8fc1e Build python only when DNN_BUILD_PYTHON is on, set PIC to on and misc changes
- cd6249e Update
- cb448e7 Fix bug of cmake
- 845d3e6 Move binary file output of Release/Debug in Windows
- 8cb4433 Set global msvc flag
- 6662958 Update .travis.yml for multiple env var
- 9841b91 Ignore .eggs
- 9288d85 Disable python 3.7 wheel build since onnx doesn't provide 3.7 wheel
- a34c395 Update ONNX submodule
- 5adeeb7 Update ONNX submodule
- 86e4686 Update ONNX submodule
- 06d7e58 Skip 32bit builds, since ONNX Runtime doesn't provide 32-bit package
- 416ea42 Fix ONNX_IMPORT
- db2bc51 Remove empty ONNX_PROTO_SRCS and headers
- c0ce646 Fix ONNX_IMPORT
- 1ef0148 Only set ONNX_BUILD_MAIN_LIB to onnx_proto
- d6a454d Add ONNX_BUILD_MAIN_LIB
- 6a0cae1 Add definitions for protobuf dll
- 6d04b45 Handle the build type
- eaeb3a4 Remove /WX in ONNX
- 262378b Use vs 2017
- eae6eda Alias for protobuf targets, workaround for protobuf 3.5.2 source
- 4f77ab4 Set protobuf git to 3.5.2 for windows ci
- 2aff852 Set conda path
- 54011e2 Move all options together
- a1f4e8b Upgrade protobuf to 3.7.1 release, set ONNX to 1.4.1 release
- fac3159 add cmake to install_require
- 0bc5e7f Only find protobuf when using system protobuf
- c5b0595 Use custom cibuildwheel in .travis.yml
- d1a03a2 Revert "Move python files a same level with c++ files"
- 16abae0 Fix appveyor submodule
- c80d8a5 Update new cibuildwheel
- afb603b Move python files a same level with c++ files
- edb0adc Update .travis.yml and appveyor.yml
- 65ac1a4 Update .travis.yml to test cibuildtools
- 80bd1e2 Update wrapper python files, fix wrong extension module name, add optional table_file argument.
- b51bf28 python package
- b0cad5b Merge pull request #51 from JDAI-CV/unify_onnx2daq
- 82ff1b5 fix wrong naming style in java wrapper
- 1f0c031 Update java wrapper for new get size
- 43e35f1 Fix wrong naming style in dnn_benchmark and apply clang-format
- 765d546 Determine output name automatically in dnn_benchmark
- cced7ed Update dnn_benchmark for to get size
- 3d21350 Disable read onnx by default
- afdec77 Revert some wrong editions in NeuralNetworksWrapper.h
- 8f05911 Fix bug in validate_onnx.py -- support non 4d input
- f2fff56 A more powerful validate_onnx.py. Only dir and dnn_retrieve_result is
- 5c514fc read onnx by default in validate_onnx.py
- 4150702 Fix wrong path in validate_onnx.py
- 896065b Add option to generate NCHW result in dnn_retrieve_result
- 76d1f8b Support multiple inputs in validate_onnx.py
- 7e7346c Import argh to handle cli args in dnn_retrieve_result
- eeeec2d Update dnn_retrieve_result to receive multiple inputs, update
- cbe6128 Update validate_onnx.py to not set output name
- fd17f22 Only delete .daq model
- 39ebfea Not set output explicitly in dnn_retrieve_result
- b0ff867 Automatically find outputs
- 5e9fed5 Apply clang-format to onnxconverter
- 42c8044 Generate code
- 0a93f71 Prepare to generate code about imm_blob_inputs_ and imm_blob_outputs_
- 1ed7bb2 Prepare to generate code about tensor_inputs_ and tensor_outputs_
- 2c00e2e Unify onnx2daq
- 60ad90b Add separated Convert, Save and GetBuf
- 3ec2638 First commit for directly onnx reading, at least it compiles
- 3191182 Update dnn_retrieve_result to receive multiple inputs, update
- 469bd79 Add DNN_ASSERT_EQ
- 3c7e7f8 Add const qualifier in some functions
- a22c6ed Support batch input in model builder
- da2565d Update AppImage
- 329df55 Update LICENSE
- c99b4a0 Upgrade onnx to v1.5.0 and protobuf to v3.7.1
- 78af419 Remove AS header in cmake
This list of changes was auto generated.
0.6.19
0.6.16
0.6.15: Move build_aar.sh to scripts/, add build_appimage.sh,
deploy appimage in ci
Please download from jcenter directly
No more extra aar file will be uploaded. Please download from jcenter directly :)