v0.20.0

Major Feature

• Move dlk to blueoil/converter #786 #847 by @kchygoe
• Move lmnet to blueoil #795 #799 #802 #818 #810 #811 #813 #822 #815 #826 #833 #824 #834 #832 #835 #839 #837 #840 #841 #838 #843 #862 by @tfujiwar @iizukak @oatawa1

Features / Refactoring

• BDD100K : Test Files (#633) by @ananno
• Remove configs using convert_weight_from_darknet (#782) by @iizukak
• Support convert without GPU (#789) by @hadusam
• Make blueoil as python module (#800) by @iizukak
• Remove scripts using darknet pretrained models (#783) by @iizukak
• Update pytablewriter 0.35 to 0.47.0 (#809) by @iizukak
• Improve PR template (#828) by @iizukak
• Remove network initialization time from benchmark (#803) by @hadusam
• Unify 2 types of config format into lmnet config (#798) by @tfujiwar
• Rename lib name (#724) by @primenumber
• Change flake8 target lmnet to blueoil (#844) by @iizukak
• Remove *.requirements.txt and Add requires to setup.cfg (#825) by @iizukak
• Remove dev.requirements.txt (#848) by @iizukak
• Unify tensor name (#829) by @primenumber
• Change for multinode training (#728) by @kurarrr
• Use integer division in pre_processor (#859) by @Joeper214
• Fix cmd to be module (#863) @kchygoe
• Update pandas 0.19.2 to 1.0.1 (#866) by @iizukak
• Update prompt-toolkit 1.0.15 to 3.0.3 (#874) by @oatawa1
• Run tests as user (#870) by @hadusam
• Use integer division on configs (#873) by @Joeper214
• Use list instead (#864) by @Joeper214
• Remove unecessary gast dependency (#879) by @Joeper214

Bug Fixes

• Fixed missing variable declaration in usb_camera_demo (#820) by @oatawa1
• Verify keypoint detection tutorial documentation (#827) by @oatawa1
• Fix benchmark script to measure elapsed time correctly (#816) by @hadusam
• Fix trial option and calculation (#807) by @hadusam
• Fix run.py import nnlib (#856) by @Joeper214
• Fix checkpoint validation (#857) by @kurarrr
• Fix some files are missing in output_template (#853) by @iizukak
• Fix python files of output_template (#871) by @hadusam
• Fix is_scalar (#869) by @primenumber

Documentation

• Include link to Blueoil repository. (#849) by @lm-lily
• Fix and update the contents of image_cls. (#850) by @lm-lily
• Update contents of Object Detection tutorial. (#851) by @lm-lily

v0.19.0

Features / Refactoring

• Remove two unrequired scripts (#736) by @tk26eng
• Parallelize test-keypoint-detection (#742) by @hadusam
• Parallelized lmnet test without GPU (#738) by @hadusam
• Remove TensorFlow's bug workaround (#748) by @tfujiwar
• Add target x86_avx to make_list of convert command (#755) by @hadusam
• Split global.tpl.h into types.h and global.tpl.h (#744) by @primenumber
• Optimize concat on depth (#756) by @primenumber
• Optimize func_Split for ChHWBCl (#759) by @primenumber
• Use ChHWBCl instead of HWChBCl if can (#729) by @primenumber
• jinja free tensor_view.h (#764) by @primenumber
• Change device buf type (#767) by @primenumber
• Remove unused config (#766) by @primenumber
• Use set instead of list to check matching by using 'in' statement (#765) by @hadusam
• Remove tensorflow-on-arm dependency (#774) by @iizukak
• Update TensorFlow 1.15.0 -> 1.15.2 (#773) by @iizukak
• Change pycocotools dependency, submodule to requirements.txt (#775) by @iizukak
• Convert prob to float before finding output class in classification prediction (#745) by @lm-jira
• Rename QTZ_binary_channel_wise_mean_scaling (#659) by @Joeper214
• Remove check-dataset-storage test (#779) by @iizukak
• Remove tox.ini (#780) by @iizukak
• Rename QTZ_binary_mean_scaling (#671) by @ Joeper214

Bug Fixes

• Ensure pool processes have closed (#726) by @joelN123
• Fix missing compile options (#760) by @primenumber
• Fix predict run.py degraded change (#770) by @kchygoe
• Close the validation dataset used in evaluate (#777) by @joelN123

v0.18.0

Features / Refactoring

• Add FusedBatchNormV3 support (#695) by @iizukak
• Update TensorFlow 1.15 and g++ 9 (#690) by @iizukak
• BDD100K : Tests Fixtures (#632) by @ananno
• Make run.py compatible with python2 (#698) by @tfujiwar
• Use logging instead of print (#705) by @hadusam
• Enhance evaluate.py to output JSON file for benchmark (#682) by @hadusam
• Enhance profile_model.py to output JSON file for benchmark (#691) by @hadusam
• Add a script to generate inference-pipline.yaml for buildkite (#686) by @fujiwar
• Use depends_on instead of wait for pipeline (#707) by @hadusam
• Add a script to generate inference-pipline.yaml for buildkite (#686) by @fujiwar
• Use symlink for output_template/python/lmnet/utils/output.py (#688) by @yd8534976
• Add details of training to json output for benchmark (#708) by @hadusam
• Optimize func_AveragePool (#713) by @primenumber
• Optimize QuantizedConv2D ApplyScalingFactor (#716) by @primenumber
• AddV2 to Add mapping (1.15.0 -> 1.14.0) (#717) by @kalpitthakkar-lm
• Change metrics json format from list to dict (#718) by @hadusam
• Add cosine learning rate scheduling (#719) by @iizukak
• Adding support for pytest with tox for single test file (#714) by @ananno

Bug Fixes

• Fix to use logger (#706) by @hadusam
• Fix a bug on initializing benchmark results in JsonOutput (#681) by @tfujiwar
• Fixed to load config file. (#678) by @odoku
• Fix pytype return for Uint16 (#721) by @kalpitthakkar-lm

Documentation

• Change tutorial in documentation to use docker image (#723) by @suttang

v0.17.1

This release include a bug fix of v0.17.

Bug Fix

• Revert "Optimize and refactor usb_camera_demo (#644)" #699 by @primenumber

v0.17.0

Major Feature

• Add keypoint detection blueoil interface (#626) by @yd8534976

Features / Refactoring

• Add docker-entrypoint.sh (#625) by @suttang
• Add runtime option to docker-compose (#637) by @suttang
• Allocate temporary buffer in initialization time (#652) by @primenumber
• Remove try wrap in blueoil train command (#650) by @suttang
• Optimize and refactor usb_camera_demo (#644) by @primenumber
• BDD100K : Segmentation (#631) by @ananno
• Remove submodule which is no longer used (#657) by @kchygoe
• Optimize batchnorm (#658) by @primenumber
• Remove unused codes (#660) by @primenumber
• Add unit test for resize pre_processor (#655) by @Joeper214
• Remove dead codes (#662) by @primenumber
• Remove func_QTZ_binary_channel_wise_mean_scaling (#586) by @primenumber
• Allocate temporary buffers in Network::init (#664) by @primenumber
• Removed some unused variable definitions (#669) by @tkng
• Change default experiment_id in blueoil/cmd/train.py (#674) by @yd8534976
• Enable OpenMP for matrix_multiplication_col3 (#675) by @primenumber
• Update numpy to 1.17.4 (#689) by @iizukak

Bug Fixes

• Fix time measurement name (#653) by @primenumber
• Fix indent and space (#654) @primenumber
• Fix include file (#667) by @primenumber
• Fix dtype of resize preprocessor (#672) by @yasumura-lm
• Fix os.rename(output, output_file_path) in blueoil/cmd/convert.py (#677) by @yd8534976

Documentation

• Add tutorials for training keypoint detection (#685) by @yd8534976

Other

• Add CODEOWNERS file (#647) by @hadusam

v0.16.0

Features / Refactoring

• Adding checker of hardware constrain for S2D (#624, #638) by @nlpng
• BDD100K : Renaming dataset classname (#628) by @ananno
• BDD100K : ObjDet class update (#630) by @ananno
• Remove the Model class (#635) by @nlpng
• Optimize quantized conv2d for arm (#559) by @primenumber
• Added image reading function for runtime (#577) by @yoya
• Add lm_single_pose_v1 network (#613) by @yd8534976
• Refactoring lmnet/executor/export.py #199 by @ruimashita
• Remove suspicious simple model (#641) by @nlpng

Bug Fixes

• Fixing to the runtime Bi-Linear Resize. (#611) by @yoya
• Fix generate lmnet config flags (quantize 1st layer and dataset prefetch) #643 by @Joeper214
• Use more proper DATA_DIR in conftest.py (#640) by @yd8534976

Documentation

• Removed lmnet/CONTRIBUTING.md. (#614) by @ed728
• Fix netlify setting (#648) by @hadusam

v0.15.0

Features / Refactoring

• Add scripts to make 32bit and 64 bit os image (#560) by @tk26eng
• Add visualize_keypoint_detection and usb camera demo (#593) by @yd8534976
• Parallelize e2e test (#605) by @suttang
• Support QUANTIZED_PACKED for debug (#622) by @primenumber
• Add time measurement functions for convert_tensor (#623) by @primenumber
• Change DLK test runner from nose2 to pytest (#604) by @hadusam
• Pad thresholds (#621) by @primenumber
• Parallelize dlk code_generation tests by architecture (#606) by @hadusam
• Add ytfaces dataset for facial landmarks detection (#610) by @yd8534976

Bug Fixes

• Fix deprecated warnings (#587) by @Joeper214
• Fix module tensorflow._api.v1.io.gfile has no attribute Remove (#609) (#616) by @yd8534976

Documentation

• Removed lmnet/docs. (#615) by @ed728

v0.14.0

Features / Refactoring

• Remove unused files (#539) by @primenumber
• Enable to save outputs of code gen test (#552) by @hadusam
• Remove func_ConcatV2 (#563) by @primenumber
• Create div2k dataset class (#547) by @suttang
• Revive segmentation test for ARM with hard_quantize=False (#557) by @primenumber
• MscocoObjectDetection can set optional classes and instantiate. (#525) by @takecore
• Move lmnet/utils/json.py to lmnet/utils/predict_output.py (#570) by @suttang
• Can use host network using docker-compose (#566) by @ruimashita
• Add buf parameter of matrix layout convertion functions (#571) by @primenumber
• Refactor predict output (#579) by @suttang
• Added deprecation warning(#581) by @twille00
• Add e2e test (#536) by @suttang
• Move requirements.txt in lmnet to the top (#561) by @iizukak
• Raise TypeError in view.run() if op_type is not supported (#540) by @yd8534976
• Refactor Makefile for DLK (#588) by @primenumber
• Collect box related functions into a box module (#513) by @yd8534976
• Refactor and generalize inputs_to_string (#590) by @primenumber
• Add FormatJoints post_processor for single-person Pose estimation (#546) by @yd8534976
• Remove blueoil.sh, blueoil_test.sh (#589) by @suttang
• Implement preserve_quantization for Prod, Shape (#585) by @primenumber
• Rename view.shape and view.shape_list (#594) by @primenumber
• Add test for binary files on aarch64 platform (#556) by @hadusam
• Implement benchmark feature in run.py (#578) by @kchygoe

Bug Fixes

• Fix dimension format for kernel (#584) by @primenumber
• Fix docker layers and stage for local development (#580) by @kchygoe
• Fix typo in cmd.py (#583) by @Joeper214
• Extend timeout of dlk tests (#576) by @hadusam
• Fix dlk/examples YAML some files. #554 by @yoya

Documentation

• Change docstring to google style (#572) by @suttang
• Fix documentation for CLI changes (#591) by @suttang
• Update the path in README #597 by @LucienZhang

v0.13.0

Features / Refactoring

• Remove duplicated method definitions in operators.py (#499) by @tkng
• Remove redundant files in in dlk (#500) by @nlpng
• Remove unnecessary code generation tests (#510) by @hadusam
• Add segmentation test of code generation for ARM/FPGA (#511) by @hadusam
• Rebuild cli structure (#514) by @suttang
• Refactoring dlk/tests/test_code_generation.py (#518) by @hadusam
• Add fpga_setup flag for setUpClass (#520) by @hadusam
• Optimize func_QTZ_linear_mid_tread_half for float output (#524) by @primenumber
• Create docker-compose.yml (#526) by @suttang
• Moved specification folder from lmnet/docs to docs (#537) by @mosesds
• Add code generation test for aarch64 (#521) by @hadusam
• Add object_keypoint_similarity metrics for single pose estimation (#517) by @yd8534976
• Reduce buffer size (#522) by @primenumber
• Remove unrequired .gitignore file (#538) by @tk26eng
• Remove Docker related files in lmnet directory. (#543) by @AlecJ
• Update Pillow to 6.2.1 (#553) by @Joeper214
• Update dlk tensorflow to 1.14.0 (#551) by @suttang

Bug Fixes

• Fix lm_resnet skip connection for unusual dimension (#504) by @nlpng
• Fix import error without FPGA_HOST environment variable (#535) by @hadusam
• Fix class name when generating metrics for Object Detection (#494) by @Joeper214

Documentation

• Remove Outdated LMnet Docs (#541) by @Joeper214

v0.12.0

Features / Refactoring

• Automatically enable Horovod when multiple GPU IDs are specified #393 by @tkng
• Remove TVM dependency #411 by @iizukak
• Remove a redundant semicolon from dlk src/func/conv2d.cpp #427 by @serihiro
• Delete absolute mode path from train executor #431 by @yasumura-lm
• Update TF version to 1.14.0, Horovod version to 0.18.1 #440 by @tkng
• Remove codes for old IP #443 by @primenumber
• Add image_size and project_name interfaces in blueoil convert #416 by @kchygoe
• Remove check-dataset-storage from 'make test-lmnet' #453 by @ruimashita
• Remove py35 target on test-lmnet #450 by @suttang
• Improve import order by isort #455 by @tkng
• Update split operator #405 by @nlpng
• Clean up tf importer #457 by @nlpng
• Well managed file descriptor and mmap-ed memory #454 by @primenumber
• Fix depth calculation for concat_on_depth #460 by @primenumber
• Remove git log output in dlk test which is unused any longer #465 by @kchygoe
• Allocate proper array size #463 by @primenumber
• Unify load_image() in lmnet #449 by @yd8534976
• Optimize batchnorm #469 by @primenumber
• Optimize matrix_shift_add #464 by @primenumber
• Update pytest version to 5.2.0 #487 by @tkng
• Layout-generalized concat_on_depth #472 by @primenumber
• Remove cURL from docker image #486 by @Sjesc
• Remove imagemagick from dockerfile #493 by @Joeper214
• Use enum class instead of struct #496 by @primenumber
• Add MSCOCO_2017 dataset for single pose #429 by @yd8534976
• Remove dlk dockerfiles which is not used any longer #501 by @kchygoe
• Remove LmThingsOnATable class #461 by @Joeper214
• Fix matrix multiplication for Arm #506 by @primenumber

Bug Fixes

• Fix debug option in the dlk #395 by @nlpng
• Fix mmap error check #447 by @primenumber
• Fix resize_nearest_neighbor support in view.py #437 by @yd8534976
• Fix typos #434 #439 #423 by @tkng, @suttang
• Fix to save image in png format correctly in lmnet/executor/export.py #456 by @ruimashita
• Remove debug display #432 by @suttang
• Use CUDA stab instead of specifying LD_LIBRARY_PATH by #425 @hadusam

Documentation

• Delete old documents #452 by @tk26eng