MMDeploy Release V1.3.1

What's Changed

🚀 Features

• Support RTMO models from mmpose by @Ben-Louis in #2597
• Add the operator nms_match for onnxruntime by @yinfan98 in #2422
• Support the deployment of CondInst from mmdet @Boomerl in #2469 and #2481
• Support HTC from mmdet by @irexyc in #2438

💥 Improvements

• Enable exporting get_simcc_maximum for simcc by @RunningLeon in #2449
• Use different stream when create handle by @irexyc in #2526
• Add the ability to build a project with PyTorch 2.0. by @Danil328 in #2553
• Add prefix mmdet to LoadImageFromNDArray by @Danil328 in #2554

🐞 Bug fixes

• Fix ort wrapper by @RunningLeon in #2560
• Fix for mask dimensions mismatch in rtmdet-ins models after exporting to ONNX by @LuukvandenBent in #2574
• Fix wrong size of mask in RTMDet-Ins post-processing by @flytocc in #2581
• Fix the onnx exportation for yoloxpose from mmpose by @Ben-Louis in #2466
• Fix building wheel on the Jetson platform by @roman-duris in #2497
• Fix "has no attribute startswith" by @Y-T-G in #2498
• Fix error when default_scope is None by @Y-T-G in #2499
• Fix pose inference on list of ndarrays by @Y-T-G in #2506

📚 Documentations

• Update the document of building mmdeploy with ncnn backend on Windows platform by @shuyaoyimei in #2476

🌐 Other

• Bump version to v1.13.1 by @RunningLeon in #2613
• Fix ci failures by @RunningLeon in #2609

New Contributors

• @shuyaoyimei made their first contribution in #2476
• @yinfan98 made their first contribution in #2422
• @Boomerl made their first contribution in #2469
• @Ben-Louis made their first contribution in #2466
• @roman-duris made their first contribution in #2497
• @LuukvandenBent made their first contribution in #2574
• @Danil328 made their first contribution in #2553
• @Y-T-G made their first contribution in #2498

Full Changelog: v1.3.0...v1.3.1

MMDeploy Release V1.3.0

What's Changed

🚀 Features

• Support DINO, Deformable DETR, Conditional DETR, DAB-DETR from mmdet by @RunningLeon in #2410
• Support PanopticFPN, MaskFormer, Mask2Former from mmdet by @RunningLeon in #2347
• Support TSM, X3D from mmaction2 by @RunningLeon in #2259
• Support onnxruntime fp16 by @RunningLeon in #2269
• Support the deployment of SMOKE from mmdet3d by @zhjunqin in #2312
• Support MobilenetV3 from mmpretrain by @Macok in #2256
• Upgrade prebuild packages to cuda-11.8, tensorrt-8.6.1, onnxruntime-1.15.1 by @RunningLeon in #2447

💥 Improvements

• Refine gemm in ort mdcn by @AllentDan in #2292
• Add time profile by @tpoisonooo in #2307
• Rename MMEdit to MMagic by @KKIEEK in #2418
• Support coco-wholebody visualization in pose_tracker python demo by @RunningLeon in #2450
• Upgrade rknn-toolkit2 version to 1.5 by @AllentDan in #2278

🐞 Bug fixes

• Prevent RuntimeError in ONNX export with empty input in ONNXNMSRotatedOp by @yhna940 in #2305
• Fix pytorch deepcopy trace error by @AllentDan in #2319
• Fix random color for Palette by @irexyc in #2356
• Add warmup for wrong profiling result by @tpoisonooo in #2339
• Fix mmdet export for coreml backend by @irexyc in #2392
• Fix rtmdet-inst sdk by @RunningLeon in #2343
• Fix ONNX Runtime Op memory leak by @grimoire in #2406
• Fix sdk error for multi-gpu execution by @irexyc in #2411
• Fix img load for super-resolution in sdk by @irexyc in #2426
• Fix ONNXRuntime Compiler Error by @flytocc in #2432
• Fix roi align symbolic function in onnx opset>=16 by @CescMessi in #2428
• Fix torch2onnx for mmdet3d by @RunningLeon in #2425
• Fix conversion with rtmdet-inst, vit, conformer on torch2.0 by @RunningLeon in #2453
• Fix mmdeploy.version_info by @xiabo123 in #2215
• Add kwargs to _scaled_dot_product_attention__tensorrt by @AllentDan in #2332

📚 Documentations

• Fix yolox-pose link by @Tau-J in #2248
• Update the c++ code in get_started docs by @GiovanniFyc in #2321
• Add Chinese documents for ort/trt by @blue-q in #2320
• Fix mmseg document by @AI-Tianlong in #2363
• Add more example in building document by @tpoisonooo in #2378

🌐 Other

• Bump version v1.3.0 by @RunningLeon in #2451
• Fix urls in CI and readthedocs by @RunningLeon in #2364
• Update QR Code by @crazysteeaam in #2371
• Add is_resize_mask cfg in inst seg by @ruisv in #2229

New Contributors

• @ruisv made their first contribution in #2229
• @xiabo123 made their first contribution in #2215
• @Tau-J made their first contribution in #2248
• @GiovanniFyc made their first contribution in #2321
• @blue-q made their first contribution in #2320
• @AI-Tianlong made their first contribution in #2363
• @zhjunqin made their first contribution in #2312
• @crazysteeaam made their first contribution in #2371
• @Macok made their first contribution in #2256
• @flytocc made their first contribution in #2432
• @CescMessi made their first contribution in #2428

Full Changelog: v1.2.0...v1.3.0

MMDeploy Release V1.2.0

What's Changed

🚀 Features

• Support YoloX-Pose by @huayuan4396 in #2184
• Support MMPretrain EfficientFormer by @huayuan4396 in #2108
• Support mscv statically linked CRT by @irexyc in #2118
• Enable read zip model in prebuild package. by @irexyc in #2185

💥 Improvements

• Reduce mmdeploy android library size by @irexyc in #2095
• Upgrade ncnn in build_ubuntu_x64_ncnn.py by @tpoisonooo in #2154
• Add coco-wholebody-hand skeleton by @huangjiyi in #2186
• Add pplcv area interpolation by @carminez88 in #2153

🐞 Bug fixes

• Fix mmseg output sync by @irexyc in #2122
• Fix export info for mmyolo by @RunningLeon in #2146
• Fix satrn onnxruntime batch inference by @irexyc in #2139
• Fix mmseg out_channels=1 by @RunningLeon in #2179
• Fix torch2onnx for pointpillars with multi-level outputs by @RunningLeon in #2210
• Fix text recog task postprocess by @irexyc in #2209
• Fix NMS Return Index by @huayuan4396 in #2196
• Fix trt api usage error by @irexyc in #2211

📚 Documentations

• Add deploee introduction by @tpoisonooo in #2120
• Update links in docs by @RunningLeon in #2114
• Update document for docker installation by @RunningLeon in #2142
• Update readme intro image and docs by @RunningLeon in #2175
• Add sdk Doxygen document. by @irexyc in #2157
• Add sdk profiler docs by @irexyc in #2189

🌐 Other

• Fix uploading assets by @RunningLeon in #2130
• Fix some ut by @RunningLeon in #2187
• Bump version to v1.2.0 by @RunningLeon in #2223

New Contributors

• @carminez88 made their first contribution in #2153
• @huangjiyi made their first contribution in #2186

Full Changelog: v1.1.0...v1.2.0

MMDeploy Release V1.1.0

What's Changed

🚀 Features

• Support SOLOv2 deployment with OpenVINO by @Li-Hongda in #1521
• Support loading openvino model from memory by @irexyc in #2027
• Add attention convertor by @AllentDan in #2064

💥 Improvements

• Rename mmcls to mmpretrain by @huayuan4396 in #2003
• Rename mmedit to mmagic by @huayuan4396 in #2061 and @AllentDan in #2092
• Add with argmax by @AllentDan in #2038
• Support opset_version 13 for ort and trt by @RunningLeon in #2071
• Swap width and height in Resize and Pad according to upstream codebases by @AllentDan in #2063

🐞 Bug fixes

• Fix retinanet wrong visualization with torch1.13 by @RunningLeon in #1977
• Fix rknn and aarch64 build ci by @irexyc in #1980
• Fix dataloader for ncnn quantization by @irexyc in #2018
• Fix with_argmax for mmseg by @RunningLeon in #2056
• Fix wrong batch index for CascadeRoiHead by @grimoire in #2078
• Fix regression test for mmocr by @AllentDan in #1789
• Fix function rewriter by @huayuan4396 in #2105

📚 Documentations

• Update SAR status by @AllentDan in #1789
• Generate supported-backends markdown table by @kaizhong2021 in #1986
• Update mmcls doc by @Renzhihan in #1978
• Update mmseg docs by @Renzhihan in #2022
• Update Repo List in README by @huayuan4396 in #2031
• Update build_from_source.md by @liuzhixin2003 in #1994
• Update mmdet.md en/cn by @Luo-Z13 in #2054
• Update Github Badge for Workflow by @huayuan4396 in #2020

🌐 Other

• Remove circleci by @RunningLeon in #1999
• Update ncnn+ppq github ci by @RunningLeon in #2058
• Add regression test workflow in github action by @RunningLeon in #2081
• Limit urllib3 for readthedocs by @RunningLeon in #2070
• Fix prebuild workflow by @RunningLeon in #2107 and #2112
• Bump version by @RunningLeon in #2094

New Contributors

• @Renzhihan made their first contribution in #1978
• @Li-Hongda made their first contribution in #1521
• @huayuan4396 made their first contribution in #2031
• @liuzhixin2003 made their first contribution in #1994
• @Luo-Z13 made their first contribution in #2054

Full Changelog: v1.0.0...v1.1.0

MMDeploy Release V1.0.0

What's Changed

🎉 Highlights

• We are excited to announce the release of MMDeploy v1.0.0. MMDeploy v1.0.0 is the first officially released version of MMDeploy 1.x, a part of the OpenMMLab 2.0 projects. Up to the release, MMDeploy 1.x supports OpenMMLab 2.0 based projects: MMCls 1.x, MMDet 3.x, MMDet3d 1.x, MMSeg 1.x, MMEdit 1.x, MMOCR 1.x, MMPose 1.x, MMAction2 1.x., MMRotate 1.x, MMYOLO.
• We switch the default branch to main from master. MMDeploy 0.x (master branch) will be deprecated and new features will only be added to MMDeploy 1.x (main branch) in future.
• We provide newly prebuilt mmdeploy packages and users can install mmdeploy through pip and download libraries from github release page for sdk inference.

🚀 Features

• Support Vacc Backend (#1679)
• Dynamically load net module to remove dependencies of mmdeploy.so (#1776)
• Sync Java apis with newly added c apis and demo (#1795)
• Add optional softmax in LinearClsHead (#1863)
• Support mmrazor (#1701)
• Support multi_scale_deform_attn trt plugin (#1844)
• Sync csharp apis with newly added c apis && demo (#1889)
• Support RTMDet and RTMPose ncnn deployment (#1857)
• Support ConformerHead from mmclassification (#1905)

🔥 Improvements

• Upgrade onnx to 1.13.0 to resolve security issue (#1793)
• Rewrite torch.cat for TensorRT when input is dynamic (#1851)
• Add github workflow based on self hosted runner (#1751)
• Optimize mmpose postprocess (#1888)
• Add rtmpose coreml support (#1902)
• Add github prebuild workflow after new release (#1852)
• Rename mmdeploy_python to mmdeploy_runtime (#1821)
• Update toolset of prebuild image (#1932)
• Update introduction image (#1937)

🐞 Bug fixes

• Fix reduction fp16 exceed on android (#1819)
• Fix unit test in test_codebase (#1790)
• Fix clip on tensorrt 8.0 (#1833)
• Fix MobileOne deployment. (#1870)
• Fix the warning method that calls the trt logger (#1850)
• Fix rtmdet-ins head (#1884)
• Fix ascend faster-rcnn (#1842)
• Fix RTMDet-Ins inference result without mask in SDK (#1867)
• Fix pose_tracker python api will raise ValueError when result has no human (#1922)
• Fix mmrotate and regression script (#1917)
• Add unified device guard (#1855)
• Fix Debian aarch64 cross compiling (#1806)
• Export mmdeploy only in monolithic build (#1798)
• Fix errors about deploying MMYOLO-OpenVINO, DETR, ConvFormer and RTMDet (#1919)

📚 Documentations

• Update get_started for installation from pypi (#1947)

🌐 Other

• Bump version to v1.0.0 (#1960)

New Contributors

• @doufengqi made their first contribution in #1165
• @RangiLyu made their first contribution in #1104
• @MambaWong made their first contribution in #1324
• @vansin made their first contribution in #1334
• @sanbuphy made their first contribution in #1398
• @DDGRCF made their first contribution in #1454
• @kota-iizuka made their first contribution in #1561
• @liuyanyi made their first contribution in #1553
• @linjie111 made their first contribution in #1700
• @Xiangxu-0103 made their first contribution in #1709
• @i-square made their first contribution in #1785
• @AndyLiu93 made their first contribution in #1679
• @LKJacky made their first contribution in #1701
• @cxiang26 made their first contribution in #1844
• @Joyphy made their first contribution in #1850
• @FishBigOcean made their first contribution in #1902

Full Changelog: v1.0.0rc3...v1.0.0

MMDeploy Release V0.14.0

What's Changed

🎉 Highlights

• We are excited to announce the release of MMDeploy v1.0.0 and v0.14.0. MMDeploy v1.0.0 is the first officially released version of MMDeploy 1.x, a part of the OpenMMLab 2.0 projects.
• We switch the default branch to main from master. MMDeploy 0.x (master branch) will be deprecated and new features will only be added to MMDeploy 1.x (main branch) in future.
• We provide newly prebuilt mmdeploy packages and users can install mmdeploy through pip and download libraries from github release pages for sdk inference.

🚀 Features

• Support TensorRT engine for onnxruntime (#1739)
• Sync csharp apis with newly added c apis && demo (#1718)
• Dynamically load net module to remove dependencies of mmdeploy.so (#1776)
• Support asf module in FPNC neck from mmocr (#1828)
• Sync Java apis with newly added c apis and demo (#1856)
• Add support for converting a inpainting model to ONNX and TensorRT (#1831)

🔥 Improvements

• Add Model::ReadConfig and simplify handle creation (#1738)
• Disable building demos when preparing prebuilt package (#1810)
• Add optional softmax in LinearClsHead (#1858)
• Add highlight version for mmdeploy 1.x (#1871)
• Optimize mmpose postprocess (#1887)
• Rename mmdeploy_python to mmdeploy_runtime (#1911)
• Update pplnn to v0.9.2 to resolve ci error (#1934)
• Upgrade openvino to 2022.3.0 (#1944)

🐞 Bug fixes

• Fix palette generation on opencv-3.x (#1778)
• Fix trtlogger instead of mm logger (#1755)
• Fix normalization to_rgb option (#1771)
• Fix missing include for gcc-10 build (#1802)
• Avoid linking static libs in monolithic build (#1797)
• Fix Debian aarch64 cross compiling (#1806)
• Add unified device guard (#1855)
• Export mmdeploy only in monolithic build (#1798)
• Fix stale workflow (#1825)
• Fix pose_tracker python api will raise ValueError when result has no human (#1923)
• Fix CascadeRoIHead export when reg_class_agnostic=True in box_head (#1900)

📚 Documentations

• Fix typo in docs/en/07-developer-guide/regression_test.md (#1907)
• Update get_started for installation from pypi (#1950)

🌐 Other

• Bump version to v0.14.0 (#1943)

New Contributors

• @yhna940 made their first contribution in #1739
• @kumailf made their first contribution in #1907
• @KKIEEK made their first contribution in #1831

Full Changelog: v0.13.0...v0.14.0

MMDeploy Release V1.0.0rc3

What's Changed

• [Fix] Fix palette generation on opencv-3.x by @lzhangzz in #1779
• [Fix] fix ut and text_det create_input by @AllentDan in #1756
• [Fix] turn -DMMDEPLOY_SHARED_LIBS=ON when building rknn by @AllentDan in #1786
• [FIX] sdk normalization to_rgb option not working(dev1.x) by @SineYuan in #1792
• [Enhancement] add Model::ReadConfig & simplify handle creation (#1738) by @lzhangzz in #1794
• [Fix] Fix tensorrt int8 by @grimoire in #1799
• [Fix] Fix component not found issue by changing class YOLOV5Head to class YOLOv5Head by @i-square in #1785
• [Fix] Fix package_tools by @irexyc in #1772
• [Fix] Avoid linking static libs in monolithic build by @lzhangzz in #1801
• [Fix] Fix missing include for gcc-10 build (#1802) by @lzhangzz in #1804
• [Fix] fix centernet by @grimoire in #1768
• [Fix] disable building demos when preparing prebuilt package by @lvhan028 in #1803
• bump version to 1.0.0rc3 by @lvhan028 in #1811

New Contributors

• @i-square made their first contribution in #1785

Full Changelog: v1.0.0rc2...v1.0.0rc3

MMDeploy Release V1.0.0rc2

What's Changed

🚀 Features

• [Feature] CodeCamp #101: Support MMDetection 3.x RTMDet model deployment on RV1126 by @Qingrenn in #1551
• [Feature] Support more MMOCR models: DBNet++, TextSnake, ABINet, MRCNN by @AllentDan in #1534
• [Feature] Support RTMDet-Ins. by @RangiLyu in #1662
• [Feature] Add coco whole-body skeleton (#1736) by @lzhangzz in #1742

💥 Improvements

• [Enhancement] [cherry-pick: #1688] remove MMDEPLOY_BUILD_SDK_CXX_API option by @lvhan028 in #1692
• [Enhancement] Add 320x320 static config for tensorrt by @hanrui1sensetime in #1689
• [Enhancement] Add mmyolo CI on mmdeploy by @hanrui1sensetime in #1650
• [Enhancement] enable TRT parse ONNX model from file by @AllentDan in #1734
• [Enhancement] check in tensorrt-fp16 deployment config for simcc by @lvhan028 in #1741

🐞 Bug fixes

• [Fix] Fix build error on cuda10.2 platform by @irexyc in #1597
• [Fix] fix log info typo of check env by @AllentDan in #1613
• [Fix] Fix typo in LetterResize transform by @lzhangzz in #1618
• [Fix] fix no test_step error in TorchWrapper by @AllentDan in #1600
• [Fix] Fixed calling miniconda script in Dockerfile by @nghiaho12 in #1629
• [Fix] fix numba numpy version not compatiable in circileci by @RunningLeon in #1611
• [Fix] Fix the performance test for RKNN by @AllentDan in #1583
• [Fix] Substitute SyncBN with BN for mmpose by @hanrui1sensetime in #1631
• [Fix] fix headless device visualize by @grimoire in #1630
• [Fix] Fix preprocess_model_config for CIFAR dataset by @hanrui1sensetime in #1659
• [Fix] fix rknn output index error in SDK by @AllentDan in #1682
• [Fix] fix lint by @triple-Mu in #1680
• [Fix] fix coreml for branch 1.x by @grimoire in #1669
• [Fix] fix mmpose regression codec sdk export by @irexyc in #1684
• [Fix] fix rknn quantization pipeline by @AllentDan in #1636
• [Fix] fix nms trt jetson 1.x by @grimoire in #1687
• [Fix] import torchvision in wrapper by @grimoire in #1702
• [Fix] fix mmaction2 visualizer by @irexyc in #1703
• [Fix] fix build_backend_model() error by @AllentDan in #1705
• [Fix] fix torch allocator resouce releasing by @AllentDan in #1711
• [Fix] fix mmocr create input from np.ndarray by @AllentDan in #1721
• [Fix] Fix double free 1.x by @grimoire in #1749
• [Fix] Fix WarpBbox and memory leak in TextRecognizer (#1745) by @lzhangzz in #1752
• [Fix] remove score filtering in rtmdet_head rewriter since it leads to error shape in batch inference by @lvhan028 in #1762
• [Fix] fix MRCNN visualization, add condition for mask resize by @AllentDan in #1767

📚 Documentations

• [Docs] update build doc for jetsons by @AllentDan in #1621
• [Docs] Correct document errors in build_from_source.md master -> 1.x by @linjie111 in #1700
• [Docs] Update docs by @Xiangxu-0103 in #1709
• [Docs] Add twitter discord medium youtube link by @vansin in #1763

🌐 Other

• [Update] Switch pip to mim in Docs and Dockerfile by @xin-li-67 in #1591
• [Refactor] Support batch inference with shape clustering (#1668) by @lzhangzz in #1733
• Cherry-pick #1439 to fix 'topk' on different devices for onnxruntime-gpu inference by @hanrui1sensetime in #1603
• cherry-pick 1608 by @RunningLeon in #1637
• Cherry-pick PR 1460 to dev-1.x by @hanrui1sensetime in #1617
• Cherry-pick PR 1366, 1370, 1460, 1577, 1598, 1601, 1612, 1626 to dev-1.x by @irexyc in #1635
• Cherry pick PR 1663, 1666, 1486 to dev-1.x by @lzhangzz in #1691
• Cherry pick PR 1625, 1715, 1725 to dev-1.x by @lzhangzz in #1726
• Cherry pick PR 1740 to dev-1.x by @irexyc in #1760
• bump version to 1.0.0rc2 by @grimoire in #1754

New Contributors

• @linjie111 made their first contribution in #1700
• @Xiangxu-0103 made their first contribution in #1709

Full Changelog: v1.0.0rc1...v1.0.0rc2

MMDeploy Release V0.13.0

What's Changed

🚀 Features

• [Feature] Support feature map output for mmsegmentation by @irexyc in #1625
• [Feature] Pose tracker C/C++/Python API&demos by @lzhangzz in #1663
• [Feature] Add coco whole-body skeleton for pose tracker demo by @lzhangzz in #1736

💥 Improvements

• [Enhancement] Support auto release note by @grimoire in #1589
• [Enhancement] suppress onnx optimizer warning by @grimoire in #1593
• [Enhancement] Better unit test. by @grimoire in #1619
• [Enhancement] update log ext in regression test by @RunningLeon in #1608
• [Enhancement] Speedup TopDownAffine by CropResizePad by @irexyc in #1486
• [Enhancement] Generate supported-backends markdown table by @kaizhong2021 in #1374
• [Enhancement] improvement(tools/scripts): pip install with user environment by @tpoisonooo in #1660
• [Enhancement] remove MMDEPLOY_BUILD_SDK_CXX_API option by @lvhan028 in #1688
• [Enhancement] Optimize C++ demos by @lzhangzz in #1715
• [Enhancement] enable TRT parse ONNX model from file by @AllentDan in #1735
• [Enhancement] Support cmake configure when system exists multiple cuda versions. by @irexyc in #1740
• [Enhancement] Loosen protobuf version criteria for onnx upgrade by @goodsong81 in #1759

🐞 Bug fixes

• [Fix] Fix example standalone build for msvc by @irexyc in #1626
• [Fix] Fix visualize api bug by @hanrui1sensetime in #1641
• [Fix] coreml by @grimoire in #1658
• [Fix] Fix several bugs in the preprocessing of single channel images by @lzhangzz in #1666
• [Fix] fix ascend by @grimoire in #1667
• [Fix]fix trt nms jetson by @grimoire in #1686
• [Fix] fix instance norm double free by @grimoire in #1748
• [Fix] Fix WarpBbox and memory leak in TextRecognizer by @lzhangzz in #1745
• [Fix] Fix build error on windows-cuda platform by @irexyc in #1598
• [Fix] Fix debug build for PoseTracker by @lzhangzz in #1601
• [Fix] Fix aligned allocations on Android by @lzhangzz in #1612
• [Fix] miniconda.sh errors in Dockerfile by @nghiaho12 in #1599
• [Fix] codebase version by @tpoisonooo in #1643
• [Fix] fix rknn output index error in SDK by @AllentDan in #1683
• [Fix] lint by @triple-Mu in #1685
• [Fix] Fixed ncnn model conversion errors in Dockerfile（no module name 'ncnn. ncnn',） by @KerwinKai in #1696
• [Fix] Fix dbnet regression test by @AllentDan in #1704
• [Fix] fix torch allocator resouce releasing by @AllentDan in #1708
• [Fix] fix stale workflow by @RunningLeon in #1713
• [Fix] Add an option to flip webcam inputs for pose tracker demo by @lzhangzz in #1725
• [Fix] Fix bug in remove_imports by @eugene123tw in #1743

📚 Documentations

• [Doc] Fix Android build doc on master branch by @hanrui1sensetime in #1620

🌐 Other

• [Refactor] Support batch inference with shape clustering by @lzhangzz in #1668
• bump version to v0.13.0 by @grimoire in #1753

New Contributors

• @kaizhong2021 made their first contribution in #1374
• @KerwinKai made their first contribution in #1696
• @eugene123tw made their first contribution in #1743
• @goodsong81 made their first contribution in #1759

Full Changelog: v0.12.0...v0.13.0

MMDeploy Release V1.0.0rc1

Features

• Add profiler for SDK (#1446)
• Support MMRotate 1.x (#1401)
• Add YOLOv5 support for RV1126 device. (#1321)
• Support Torch JIT Modulated Deformable Conv (#1536)
• Support SOLO deployment with OpenVINO (#1454)
• Support TVM (#1531)
• Support Rotated RTMDet deployment (#1553)

Improvements

• Refactor SDK registry (#1368)
• Avoid copying dense arrays in Python API (#1349)
• Update dockerfile pip source (#1484)
• Add rknn device check (#1363)
• cherry-pick: Decouple preprocess operation and transformation (#1353)
• sync #1493 to support TorchAllocator as TensorRT Gpu Allocator and fix DCNv2 tensorrt plugin error (#1519)
• Add md link check github action (#1320)
• Remove cudnn dependency for transform 'mmaction2::format_shape' (#1509)
• Refactor rewriter context for MMRazor (#1483)
• Add is_batched argument to pipeline.json (#1528)
• Refactor Backend Manager(#1515)

Bug fixes

• Support ONNXRuntime-1.13 (#1407)
• Fix det->pose demo (#1419)
• Fix MMOCR import typing error (#1497)
• Fix requirements of MMEditing (#1496)
• recovery mmdet layers (#1526)
• Fix squeeze export and unsqueeze pass for opset 13 (#1538)
• Fix 'cannot seek vector iterator' in debug windows build (#1543)
• Fix API build error in readthedocs (#1567)
• Fix unittest and suppress warning (#1552)
• Rename 'forward_test' to 'predict' (#1561)
• Fix onnx2ncnn.cpp bugs (#1518)

Document

• Add mmaction2 & coreml index for readthedocs (#1542)

Contributors

@lzhangzz @lvhan028 @PeterH0323 @AllentDan @irexyc @RunningLeon @grimoire @hanrui1sensetime @pppppM @antoszy @DDGRCF @kota-iizuka @liuyanyi