MMDeploy Release V0.5.0

[lvhan028]

Highlight

• Provide prebuilt packages since v0.5.0
• Decouple pytorch2onnx and onnx2backends
• Support text detection models PANet, PSENet and DBNet, with CUDA accelerated postprocessing in SDK
• Support MMRotate

Features

• Add prebuild tools ([Prebuild] Add prebuilt TensorRT build number #545, [Refactor][tools] Add prebuild tools. #347)
• Experimental executor support in SDK ([Feature] New pipeline & executor for SDK #497)
• Support ViT on ncnn (improvement(ViT): use Crop to subtitude Gather #477, feat(codebase/cls): support vision_transformer #403)
• Support LiteHRNet on ncnn ([Enhancement] Support litehrnet for ncnn #316)
• Support more text detection models PANet, PSENet and DBNet, with CUDA accelerated postprocessing in SDK (More ocr models #446, [Feature] Support DBNet, PANet and PSENet for SDK, with GPU aided post-processing #526, change resolution and fix regression test for mmocr #534)
• Add C# API for SDK (add csharp support #388, fix C# API build option #535)
• Support ncnn quantization (feat(tools/deploy.py): support ncnn quantization #476)
• Support RepPoints on TensorRT ([Feature] Support reppoints TensorRT #457)
• Support MMRotate on ONNX Runtime and TensorRT (Add roi_align_rotated onnxruntime op #277, Add nms_rotated ort op #312, [Enhancement] Support RotatedRetinaNet TensorRT #422, mmrotate sdk module #450, Support single stage rotated detector in MMRotate #428, [Feature] Support two stage rotated detector in MMRotate #473)
• Support MMRazor ([Feature] Add mmrazor support #220, fix import mmrazor error in mmcls test #467)

Improvements

• Remove spdlog manual installation but still keep it as an option ([Fix] support latest spdlog #423, [Fix] Fix recent build problems #544)
  Users can turn on the following option to use the external spdlog

cmake .. -DMMDEPLOY_SPDLOG_EXTERNAL=ON

• Add SDK python demos (add sdk python demo #554)
• Add ONNX passes support ([Enhancement] Add ONNX passes support #390)
• Decouple pytorch2onnx and onnx2backends ([Refactor][API2.0] Api refactor2.0 #529, [Fix] Fix tools/checkenv.py #540)
• Add scripts and configs to test metrics of deployed model with all inference backend ([Enhancement] Improve MMDeploy Regression test  #425, [Feature] Regression test for mmdeploy #302, Fix test ncnn failure in regression test #551, [RegressionTest] Add mmrotate regression test #542)
• Support MDCN and DeformConv TensorRT FP16 ([Enhancement] Add TensorRT fp16 support for MDCN #503, [Enhancement] Support DeformConv TensorRT fp16 #468)
• Add interactive build script for Linux and NVIDIA platform (MMdeploy build & install script + prerequisites #399)
• Optimize global average pooling when exporting ONNX (Rewrite mmcls gap #478)
• Refactor onnx2ncnn, add test cases and simplify code (refactor(onnx2ncnn): add test case and simplify code #436)
• Remove expand operation from mmdet rewrite ([Enhance] remove expand from mmdet rewriter #371)

Bug fixes

• Update CMake scripts to fix building problems ([Fix] Fix recent build problems #544, fix(demo/csrc/CMakeLists.txt): codebase build option #553)
• Make ONNXRuntime wrapper work both for cpu and cuda execution (fix mmdet ONNXRuntime cuda test bug #438, changing the onnxwrapper script for gpu issue #532)
• Fix PSPNet-TorchScript conversion error (fix pspnet torchscript conversion #538)
• Resolve the incompatible issue when upgrading MMPose from v0.25.0 to v0.26.0 (fix: compatable with mmpose v0.26 #518, compat mmpose v0.26 (sdk) #527)
• Fix mismatched device issue when testing Mask R-CNN deployed model (Fix mask test with mismatched device #511)
• Remove redundant resize in mmseg EncoderDecoder rewrite (fix mmseg twice resize #480)
• Fix display bugs on headless devices ([Fix] Fix display bugs for windows #451)
• Fix MMDet3D pillarencode deployment failure ([Fix]fix pillarencode #331)
• Make the latest spdlog compatible ([Fix] support latest spdlog #423)
• Fix CI ([Fix] Fix windows unit test #462, fix(zh_CN): doc link #447, [Fix] Limit mmpose requirement #440, [Fix] Fix ci #426, [Fix] Fix ci in master branch #441)
• Fix a bug that causes exporting to onnx failed with static shape and batch size > 1 (fix: bbox_nms not onnxizing if batch size > 1 #501)
• Make --work-dir default to $pwd in tools/deploy.py ([Fix] set optional arg a default value #483)

Documentation

• Fix user document errors, reorganize them, update REAME and rewrite the GET_STARTED chapters (docs(docs/zh_cn): update doc and link checker #418, [Docs] adding new backends when using MMDeploy as a third package #482, [docs] Fix typo in tutorial #509, [Docs] reorganize the documents in English and update the contents based on v0.5.0 status #531, docs(do_regression_test): fix-typo #547, [Docs] Update doc for dependency #543)
• Rewrite the get_started for Jetson platforms (Update: Optimize document #484, Jetson install guide additions, torch install steps updated, version comments added. #449, [Docs] Minor fixes and translation of installation tutorial for Jetson #415, [Doc] Improve Jetson tutorial install doc #381)
• Fix APIs rendering failure in readthedocs ([Docs] Fix API documentation #443)
• Remove '' in API docstring (Fix docstring format #495)
• More tutorials in Chinese are checked in - Tutorial 05: ONNX Model Editing and Tutorial 04: onnx custom op (Tutorial 04: onnx custom op #508, Tutorial 05: ONNX Model Editing #517)

Contributors

@sanjaypavo @PeterH0323 @tehkillerbee @zytx121 @triple-Mu @zhiqwang @gyf304 @lakshanthad @Dchaoqun @zhouzaida @NagatoYuki0943 @VVsssssk @irexyc @RunningLeon @hanrui1sensetime @lzhangzz @grimoire @tpoisonooo @AllentDan @SingleZombie

---

This discussion was created from the release MMDeploy Release V0.5.0.