v0.11.0

Major Features

• Update rbf and preloader. 100MHz -> 150MHz. #381 by @lm-chinhtran @n-nez
• Optimize dlk runtime #390 by @primenumber

Features / Refactoring

• Update trainer parameters for checkpoint saving #372 by @Joeper214
• Support resize_nearest_neighbor op #378 by @yd8534976
• Custom TFDS Builder / Executor Script #392 by @tfujiwar
• Port Packer class to Python #402 by @tkng
• Runtime post process implementation #253 by @yoya
• Add unit tests for Packer class #403 by @tkng
• Remove ONNX code in DLK directory #409 by @iizukak
• Remove base_module.py #406 by @tkng

Bug Fixes

• Show only rank 0 progbar #385 by @tkng
• use tf.cast instead of tf.to_int32, etc #394 by @tkng
• Fix round #400 by @yd8534976
• call flush method of a writer after the writer wrote summary data #384 by @tkng
• Remove unnecessary summary from export.py and profile_model.py #383 by @yd8534976
• Fix input size bug of rbf. #381 by @n-nez @trivedijay

Documentation

• Change CI Badge Branch to master #382 by @iizukak
• Fix reference doc links #388 by @Joeper214
• Add FAQ section and improve installation and tutorial section #389 by @OscarWang114

v0.10.0

Major Features

• Remove Jenkins #376 by @hadusam
• More optimization for x86 AVX #373 by @primenumber

Features / Refactoring

• Implemented TFDSObjectDetection class. #370 by @tfujiwar
• Remove redundant part of time measurement reporting. #360 by @ed728

Bug Fixes

• Remove broken feature from Makefile.tpl #344 by @yd8534976
• Fix queue compatibilty in usb camera demo #379 by @Joeper214

Documentation

• Documentation for time measurement. #358 @ed728
• Fix wrong spelling and syntax. #375 by @ed728

v0.9.0

Major Features

• Optimize for x86 AVX #365 by @primenumber
• Optimize for AArch32/AArch64 #353 #359 by @primenumber

Features / Refactoring

• Remove unused pygame library #363 by @Joeper214
• Introduce Buildkite to CI #357 by @hadusam
• Enable to test blueoil part by each tasks #356 by @hadusam
• Improve TFDS Dataset class and iterator #354 by @tfujiwar
• Enable Resume Training for Object Detection #352 by @Joeper214
• Enable to run specified tests of lmnet by selecting tox env #340 by @hadusam
• Show progressbar during training #336 by @tsawada

Bug Fixes

• Enable to kill docker process by Ctrl-c #351 by @hadusam
• Enable to load config file under lmnet dir #342 by @ruimashita
• Fix unstable blueoil tests #341 by @hadusam

v0.8.0

Major Features

• Add new accelerator IP TCA IPv2. #321 by @n-nez and @antonionevado @primenumber @iizukak and @nlpng @ed728 @LMSuto @lm-tajimi

• Add a dataset class for loading Tensorflow Datasets. #276 by @tfujiwar

• Implement TensorView. #248 by @primenumber

Features / Refactoring

• lmnet: Add cityscapes loader .#300 by @Joeper214 @tsawada
• Delete CustomSpaceToDepth backword function. #294 by @ruimashita
• Enable to read Blueoil config from Google Cloud Storage. #280 by @tfujiwar
• Add alias for gather. #345 by @yd8534976

Bug Fixes

• Update circleci conf for firebase-tool required version. #324 @kchygoe
• fix to wait ssh daemon enabled. #327 by @hadusam
• Change DLK test to use correct FPGA files as same as Blueoil's output. #306 by @hadusam
• Change the runtime Resize behavior to Pillow-6.0.0. #299 by @yoya
• Fix resize_with_gt_boxes() . #297 by @yd8534976
• Fix pass_lookup(). #347 by @yd8534976
• Fixed unsigned to signed integer conversion issue. #339 by @ed728
• Removed unused generate_consts. #320 by @ed728

Documentation

• Fix site uri blue-oil.org to blueoil.org. #301 by @suttang

v0.7.0

Features / Refactoring

• Update Pillow Library. #255 by @iizukak
• Enable to test DLK on local DE10-Nano. #261 by @hadusam
• Enable data prefetch for hyper-param tuner #262 by @nlpng
• Efficiency improvement of Docker build. #273 by @odoku

Bug Fixes

• Fix config overwrite by check & delete if destination exists. #258 by @Joeper214
• Add try catch to ensure queue import compatibility. #264 by @Joeper214
• Add input data shaping for images_placeholder. #265 by @wtnb93
• Fix the error of pep8-dlk. #268 by @hadusam

Documentation

• Update docs for CUDA10. #271 by @ruimashita

v0.6.0

Major Features

• Hyper Parameter Turning in lmnet. #209 #215 @nlpng
• Motion JPEG Server for demo. #241 @ruimashita

Features / Refactoring

• Use QuantizedPacked for kernel. #226 @primenumber
• Refactoring lmnet/executor/export.py run() . #227 @ruimashita
• Optimize Bilinear postprocess. #228 @ruimashita
• Fix typo: s/pyhon/python/. #237 @tsawada
• Accelerate object detection anchors generator in early steps . #250 @yd8534976

Bug Fixes

• Fix MaxPool implementation #238 @tvlenin
• Fix re-training failure and add test for re-training. #246 @hadusam

Documentation

• Update TensorFlow, CUDA and OepnCV version #239 @iizukak

v0.5.0

Major Features

• Update TensorFlow Version (1.4.1 -> 1.13.1). issue: #61 PR: #202 @lm-jira
• Add faster semantic segmentation network (LMBiSeNet) to lmnet. #225 #212 #211 @ruimashita

Features / Refactoring

• Save blueoil training config yaml when convert. issue: #197 PR: #198 @odoku
• Create Makefile and unify commands #194 @kchygoe
• Change executor/profile_model.py to delete training only operation from profile output #207 @ruimashita
• Add support of LeakyReLU to DLK #214 @nlpng
• Avoid importing entire pyplot #218 @tsawada
• Implement packed data type #217 @primenumber
• Create lmnet config with tempfile #219 @tsawada
• Export inference test data and add FUNC_TIME_MEASUREMENT macro for test & debug #184 @ruimashita
• Use gfile to read/write configs #223 @tsawada
• Semantic segmentation python demo #155 @nlpng
• Remove unused dlk/.github/CODEOWNERS file #200 @ruimashita
• Fix document structure to clean the directory #216 @kchygoe

Bug Fixes

• Fix branch condition in generic quantized matmul issue: #152 PR: #222 @primenumber
• Change default model in usb camera demo #230 @Joeper214

v0.4.1

Bug Fixes

• Fix inference run.py raise SystemError on object detection task. issue: #203, PR: #204 @ruimashita
• Add backwards compatibility for inference python scripts. PR: #201 @Joeper214

v0.4.0

Major Features

• Support Semantic Segmentation task #176 #177 #180 #189 by @iizukak
• Enable to inference on Nvidia GPU (using tensorflow) #144 by @Joeper214
• Enable to inference on AArch64 #146 by @lm-konda

Features / Refactoring

• Add Blueoil python CLI #105 by @odoku
• Improve performance by utilizing cache locality #131 by @lm-konda
• Add timer to high cost part #151 by @lm-konda
• Implement SimpleDatasetReader #166 by @ruimashita
• Add General Purpose Color Map #188 by @iizukak
• Implement image resize for inference #9 by @yoya
• implement DatasetIterator class #122 by @tkng
• Remove redundant variable from generic batch normalization code #134 by @lm-konda
• Remove redundant -g option in Makefile #156 by @iizukak

Bug Fixes

• Add bug-fixed version of quantized convolution IP with kn2row and tiling #130 by @tkclimb
• Fix overflowError: Python int too large to convert to C long #170 by @nlpng
• Fix loop boundaries in vectorized batch normalization #172 by @n-nez

Documentation

• Add Semantic Segmentation Tutorial #174 by @iizukak

v0.3.0

Features

• Adding support for ResNet. #65 @nlpng
• Optimizer selector. #28 @lm-konda @iizukak
• Added pattern matching and new optimization passes #53 @antonionevado @nlpng @n-nez
• Improve kernel transform. #73 @lm-konda
• xml parser use xml.etree.ElementTree instead of BeautifulSoup4 #87 #107 @ruimashita @tkng

Bug fixes

• Redundant or bug branch instruction (not for FPGA version). #100 @lm-konda
• Modify tie behavior of top-k accuracy calculation. #58 @joelN123
• The implementation of sqrt was actually rsqrt. #116 @lm-konda
• Fix non-existing path. #145 @djKooks

Refactoring

• Some Simple refactoring. #76 @lm-konda
• Improve max pooling implementation. #137 @lm-konda
• Improve average pooling implementation. #136 @lm-konda
• Remove unsused variables and memory initilization. #135 @lm-konda
• Update pyyaml version for security. #142 @ruimashita

Documentation

• Add favicon for documentation site. #129 @toohsk
• Add Object Detection Tutorial. #33 @iizukak