MMDetection3D v1.4.0 Release

Highlights

• Support the training of DSVT in projects (#2738)
• Support Nerf-Det in projects (#2732)
• Refactor Waymo dataset (#2836)

New Features

• Support the training of DSVT in projects (#2738)
• Support Nerf-Det in projects (#2732)
• Support MV-FCOS3D++ (#2835)
• Refactor Waymo dataset (#2836)

Improvements

• Support PGD (front-of-view / multi-view) on Waymo dataset (#2835)
• Release new Waymo-mini for verify some methods or debug quickly (#2835)

Bug Fixes

• Fix MinkUNet and SPVCNN some wrong configs (#2854)
• Fix incorrect number of arguments in PETR (#2800)
• Delete unused files in mmdet3d/configs (#2773)

Contributors

A total of 5 developers contributed to this release.

@sunjiahao1999, @WendellZ524, @Yanyirong, @JingweiZhang12, @Tai-Wang

New Contributors

• @WendellZ524 made their first contribution in #2800
• @Yanyirong made their first contribution in #2732

Full Changelog: v1.3.0...v1.4.0

MMDetection3D v1.3.0 Release

Highlights

• Support CENet in projects (#2619)
• Enhance demos with new 3D inferencers (#2763)

New Features

• Support CENet in projects (#2619)

Improvements

• Enhance demos with new 3D inferencers (#2763)
• Add BEV-based detection pipeline in nuScenes dataset tutorial (#2672)
• Add the new config type of Cylinder3D in mmdet3d/configs (#2681)
• Update New Config Type (#2655)
• Update the QR code in README.md (#2703)

Bug Fixes

• Fix the download script of nuScenes dataset (#2660)
• Fix circleCI and GitHub workflow configuration (#2652)
• Fix the version of Open3D in requirements (#2633)
• Fix unused files in mmdet3d/configs (#2773)
• Fix support devices in FreeAnchor3DHead (#2769)
• Fix readthedocs building and link (#2739, #2650)
• Fix the pitch angle bug in LaserMix (#2710)

Contributors

A total of 7 developers contributed to this release.

@sunjiahao1999, @Xiangxu-0103, @ZhaoCake, @LRJKD, @crazysteeaam, @wep21, @zhiqwang

New Contributors

• @wep21 made their first contribution in #2660
• @zhiqwang made their first contribution in #2650
• @1uciusy made their first contribution in #2672
• @crazysteeaam made their first contribution in #2703
• @ZhaoCake made their first contribution in #2681
• @LRJKD made their first contribution in #2769

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

MMDetection3D v1.2.0 Release

Highlights

• Support New Config Type in mmdet3d/config (#2608)
• Support the inference of DSVT in projects (#2606)
• Support downloading datasets from OpenDataLab using mim (#2593)

New Features

• Support New Config Type in mmdet3d/config (#2608)
• Support the inference of DSVT in projects (#2606)
• Support downloading datasets from OpenDataLab using mim (#2593)

Improvements

• Enhanced visualization in interactive form (#2611)
• Update README.md and Model Zoo (#2599, #2600)
• Speed up S3DIS data preparation (#2585)

Bug Fixes

• Remove PointRCNN in benchmark training (#2610)
• Fix wrong indoor detection visualization (#2625)
• Fix MinkUNet download link (#2590)
• Fix the formula in the readthedocs (#2580)

Contributors

A total of 5 developers contributed to this release.

@sunjiahao1999, @Xiangxu-0103, @JingweiZhang12, @col14m, @zhulf0804

New Contributors

• @col14m made their first contribution in #2585
• @zhulf0804 made their first contribution in #2600

Full Changelog: v1.1.1...v1.2.0

MMDetection3D v1.1.1 Release

Highlights

• Support TPVFormer in projects (#2399, #2517, #2535)
• Support the training of BEVFusion in projects (#2546)
• Support lidar-based 3D semantic segmentation benchmark (#2530, #2559)

New Features

• Support TPVFormer in projects (#2399, #2517, #2535)
• Support the training of BEVFusion in projects (#2558)
• Support lidar-based 3D Semantic Segmentation Benchmark (#2530, #2559)
• Support test-time augmentation for Segmentor (#2382)
• Support Minkowski ConvModule and Residual Block (#2528)
• Support the visualization of multi-view images in multi-modal methods (#2453)

Improvements

• Upload checkpoints and training log of PETR (#2555)
• Replace np.float by default float in segmentation evaluation (#2527)
• Add docs of converting SemanticKITTI datasets (#2515)
• Support different colors for different classes in visualization (#2500)
• Support tensor-like operations for BaseInstance3DBoxes and BasePoint(#2501)
• Add information of LiDAR Segmentation in NuScenes annotation files
• Provide annotation files of datasets generated offline (#2457)
• Refactor document structure (#2429)
• Complete typehints and docstring (#2396, #2457, #2468, #2464, #2485)

Bug Fixes

• Fix the bug of abnormal loss when training SECOND in Automatic mixed precision(AMP) mode (#2452)
• Add a warning in function post_process_coords in mmdet3d/dataset/convert_utils.py (#2557)
• Fix invalid configs (#2477, #2536)
• Fix bugs of unit test (#2466)
• Update local-rank argument in test.py for pytorch 2.0 (#2469)
• Fix docker file (#2451)
• Fix demo and visualization (#2453)
• Fix SUN RGB-D data converter (#2440)
• Fix readthedocs building (#2459, #2419, #2505, #2396)
• Fix CI (#2445,#2424)
• Loose the version restriction of numba (#2416)

Contributors

A total of 12 developers contributed to this release.

@sunjiahao1999, @Xiangxu-0103, @JingweiZhang12, @chriscarving, @jaan1729, @pd-michaelstanley, @filaPro, @kabouzeid, @A-new-b, @lbin, @Lum1104, @pd-michaelstanley

New contributors

• @A-new-b made their first contribution in #2485
• @lbin made their first contribution in #2557
• @Lum1104 made their first contribution in #2464
• @jaan1729 made their first contribution in #2451
• @pd-michaelstanley made their first contribution in #2442
• @chriscarving made their first contribution in #2396

Full Changelog: v1.1.0...v1.1.1

MMDetection3D v1.1.0 Release

We are excited to announce the release of MMDetection3D 1.1.0 as a part of the OpenMMLab 2.0 project! Compared with version 1.0.0, MMDetection3D 1.1.0 introduces an updated framework structure for the core package and a new section called Projects. Specifically, we have significantly refactored the core package's code to make it clearer, more comprehensible, and disentangled. The new Projects section serves as an essential addition to MMDetection3D and supports flexible code contribution without strict code requirements, enabling faster integration of state-of-the-art models and features. To help users migrate from version 1.0.0 to 1.1.0 as smoothly as possible, we have prepared a migration guide. Any questions about migration are welcome to be posted on issue。

Highlights

• Support several popular LiDAR segmentation methods:
  • Cylinder3D (#2291, #2344, #2350)
  • MinkUnet (#2294, #2358)
  • SPVCNN (#2320，#2372)
  • PolarMix augmentation (#2265)
  • LaserMix augmentation (#2302)
• Support TR3D detector in projects (#2274)
• Support DETR3D in projects (#2173)
• Support the inference of BEVFusion in projects (#2175)

New Features

• Support Cylinder3D (#2291, #2344, #2350)
• Support MinkUnet (#2294, #2358)
• Support SPVCNN (#2320，#2372)
• Support TR3D detector in projects (#2274)
• Support the inference of BEVFusion in projects (#2175)
• Support DETR3D in projects (#2173)
• Support PolarMix and LaserMix augmentations (#2265,#2302)
• Support loading annotation of panoptic segmentation (#2223)
• Support panoptic segmentation metric (#2230)
• Add inferencer for LiDAR-based, monocular and multi-modality 3D detection (#2208, #2190, #2342)
• Add inferencer for LiDAR-based segmentation (#2304)

Improvements

• Support lazy_init for CBGSDataset (#2271)
• Support generating annotation files for test set on Waymo (#2180)
• Enhance the support for SemanticKitti (#2253, #2323)
• File I/O migration and reconstruction (#2319)
• Support format_only option for Lyft, NuScenes and Waymo datasets (#2333, #2151)
• Replace np.transpose with torch.permute to speed up (#2277)
• Allow setting local-rank for pytorch 2.0 (#2387)

Bug Fixes

• Fix the problem of reversal of length and width when drawing heatmap in CenterFormer (#2362)
• Deprecate old type alias due to the new version of numpy (#2339)
• Lose trimesh version requirements to fix numpy random state (#2340)
• Fix the device mismatch error in CenterPoint (#2308)
• Fix the bug of visualization when there are no boxes (#2231)
• Fix bug of counting ignore index in IOU in segmentation evaluation (#2229)

Contributors

A total of 19 developers contributed to this release.
@xizaoqu, @ZLTJohn, @SekiroRong, @yechenzhi, @shufanwu, @chriscarving, @vansin, @triple-Mu, @404Vector, @filaPro, @sunjiahao1999, @Ginray, @Xiangxu-0103, @JingweiZhang12, @DezeZhao, @ZCMax, @roger-lcc, @Tai-Wang, @ZwwWayne

New contributors

• @xizaoqu made their first contribution in #2385
• @yechenzhi made their first contribution in #2385
• @DezeZhao made their first contribution in #2362
• @chriscarving made their first contribution in #2396

Full Changelog: v1.1.0rc3...v1.1.0

MMDetection3D v1.1.0rc3 Release

Highlights

• Support CenterFormer in projects (#2175)
• Support PETR in projects (#2173)

New Features

• Support CenterFormer in projects (#2175)
• Support PETR in projects (#2173)
• Refactor ImVoxelNet on SUN RGB-D into mmdet3d v1.1 (#2141)

Improvements

• Remove legacy builder.py (#2061)
• Update customize_dataset documentation (#2153)
• Update tutorial of LiDAR-based detection (#2120)

Bug Fixes

• Fix the configs of FCOS3D and PGD (#2191)
• Fix numpy's ValueError in update_infos_to_v2.py (#2162)
• Fix parameter missing in Det3DVisualizationHook (#2118)
• Fix memory overflow in the rotated box IoU calculation (#2134)
• Fix lidar2cam error in update_infos_to_v2.py for nus and lyft dataset (#2110)
• Fix error of data type in Waymo metrics (#2109)
• Update bbox_3d information in cam_instances for mono3d detection task (#2046)
• Fix label saving of Waymo dataset (#2096)

Contributors

A total of 10 developers contributed to this release.

@SekiroRong, @ZLTJohn, @vansin, @shanmo, @VVsssssk, @ZCMax, @Xiangxu-0103, @JingweiZhang12, @Tai-Wang, @lianqing11

New Contributors

• @shanmo made their first contribution in #2118
• @ZLTJohn made their first contribution in #2162
• @SekiroRong made their first contribution in #2175

Full Changelog: v1.1.0rc2...v1.1.0rc3

MMDetection3D V1.0.0rc6 Release

New Features

• Add Projects/ folder and the first example project (#2082)

Improvements

• Update Waymo converter to save storage space (#1759)
• Update model link and performance of CenterPoint (#1916)

Bug Fixes

• Fix GPU memory occupancy problem in PointRCNN (#1928)
• Fix sampling bug in IoUNegPiecewiseSampler (#2018)

Contributors

A total of 6 developers contributed to this release.

@oyel, @zzj403, @VVsssssk, @Tai-Wang, @tpoisonooo, @JingweiZhang12, @ZCMax, @ZwwWayne

Full Changelog: v1.0.0rc5...v1.0.0rc6

MMDetection3D V1.1.0rc2 Release

Highlights

• Support PV-RCNN
• Speed up evaluation on Waymo dataset

New Features

• Support PV-RCNN (#1597, #2045)
• Speed up evaluation on Waymo dataset (#2008)
• Refactor FCAF3D into the framework of mmdet3d v1.1 (#1945)
• Refactor S3DIS dataset into the framework of mmdet3d v1.1 (#1984)
• Add Projects/ folder and the first example project (#2042)

Improvements

• Rename CLASSES and PALETTE to classes and palette respectively (#1932)
• Update metainfo in pkl files and add categories into metainfo (#1934)
• Show instance statistics before and after through the pipeline (#1863)
• Add configs of DGCNN for different testing areas (#1967)
• Remove testing utils from tests/utils/ to mmdet3d/testing/ (#2012)
• Add typehint for code in models/layers/ (#2014)
• Refine documentation (#1891, #1994)
• Refine voxelization for better speed (#2062)

Bug Fixes

• Fix loop visualization error about point cloud (#1914)
• Fix image conversion of Waymo to avoid information loss (#1979)
• Fix evaluation on KITTI testset (#2005)
• Fix sampling bug in IoUNegPiecewiseSampler (#2017)
• Fix point cloud range in CenterPoint (#1998)
• Fix some loading bugs and support FOV-image-based mode on Waymo dataset (#1942)
• Fix dataset conversion utils (#1923, #2040, #1971)
• Update metafiles in all the configs (#2006)

Contributors

A total of 12 developers contributed to this release.

@vavanade, @oyel, @thinkthinking, @PeterH0323， @274869388, @cxiang26, @lianqing11, @VVsssssk, @ZCMax, @Xiangxu-0103, @JingweiZhang12, @Tai-Wang

New Contributors

• @PeterH0323 made their first contribution in #2065
• @cxiang26 made their first contribution in #1965
• @vavanade made their first contribution in #2031
• @oyel made their first contribution in #2017
• @thinkthinking made their first contribution in #2026
• @274869388 made their first contribution in #1973

Full Changelog: v1.1.0rc1...v1.1.0rc2

MMDetection3D V1.1.0rc1 Release

Highlights

• Support a camera-only 3D detection baseline on Waymo, MV-FCOS3D++

New Features

• Support a camera-only 3D detection baseline on Waymo, MV-FCOS3D++, with new evaluation metrics and transformations (#1716)
• Refactor PointRCNN in the framework of mmdet3d v1.1 (#1819)

Improvements

• Add auto_scale_lr in config to support training with auto-scale learning rates (#1807)
• Fix CI (#1813, #1865, #1877)
• Update browse_dataset.py script (#1817)
• Update SUN RGB-D and Lyft datasets documentation (#1833)
• Rename convert_to_datasample to add_pred_to_datasample in detectors (#1843)
• Update customized dataset documentation (#1845)
• Update Det3DLocalVisualization and visualization documentation (#1857)
• Add the code of generating cam_sync_labels for Waymo dataset (#1870)
• Update dataset transforms typehints (#1875)

Bug Fixes

• Fix missing registration of models in setup_env.py (#1808)
• Fix the data base sampler bugs when using the ground plane data (#1812)
• Add output directory existing check during visualization (#1828)
• Fix bugs of nuScenes dataset for monocular 3D detection (#1837)
• Fix visualization hook to support the visualization of different data modalities (#1839)
• Fix monocular 3D detection demo (#1864)
• Fix the lack of num_pts_feats key in nuscenes dataset and complete docstring (#1882)

Contributors

A total of 10 developers contributed to this release.

@ZwwWayne, @Tai-Wang, @lianqing11, @VVsssssk, @ZCMax, @Xiangxu-0103, @JingweiZhang12, @tpoisonooo, @ice-tong, @jshilong

New Contributors

• @ice-tong made their first contribution in #1838

Full Changelog: v1.1.0rc0...v1.1.0rc1

MMDetection3D V1.0.0rc5 Release

New Features

• Support ImVoxelNet on SUN RGB-D (#1738)

Improvements

• Fix the cross-codebase reference problem in metafile README (#1644)
• Update the Chinese documentation about getting started (#1715)
• Fix docs link and add docs link checker (#1811)

Bug Fixes

• Fix a visualization bug that is potentially triggered by empty prediction labels (#1725)
• Fix point cloud segmentation visualization bug due to wrong parameter passing (#1858)
• Fix Nan loss bug during PointRCNN training (#1874)

Contributors

A total of 11 developers contributed to this release.

@ZwwWayne, @Tai-Wang, @filaPro, @VVsssssk, @ZCMax, @Xiangxu-0103, @holtvogt, @tpoisonooo, @lianqing01, @TommyZihao, @aditya9710

New Contributors

• @tpoisonooo made their first contribution in #1614
• @holtvogt made their first contribution in #1725
• @TommyZihao made their first contribution in #1778
• @aditya9710 made their first contribution in #1889

Full Changelog: v1.0.0rc4...v1.0.0rc5