{"id":13442878,"url":"https://github.com/CVMI-Lab/SparseKD","last_synced_at":"2025-03-20T15:31:19.164Z","repository":{"id":61352164,"uuid":"539963500","full_name":"CVMI-Lab/SparseKD","owner":"CVMI-Lab","description":"(NeurlPS 2022) Towards Efficient 3D Object Detection with Knowledge Distillation","archived":false,"fork":false,"pushed_at":"2023-07-26T03:52:46.000Z","size":535,"stargazers_count":110,"open_issues_count":2,"forks_count":10,"subscribers_count":5,"default_branch":"master","last_synced_at":"2024-10-28T06:00:04.823Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/CVMI-Lab.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null}},"created_at":"2022-09-22T12:11:50.000Z","updated_at":"2024-09-02T13:14:06.000Z","dependencies_parsed_at":"2024-01-18T14:41:23.225Z","dependency_job_id":"7ea21595-196e-42e5-89d6-ca9f77c85762","html_url":"https://github.com/CVMI-Lab/SparseKD","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CVMI-Lab%2FSparseKD","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CVMI-Lab%2FSparseKD/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CVMI-Lab%2FSparseKD/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CVMI-Lab%2FSparseKD/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/CVMI-Lab","download_url":"https://codeload.github.com/CVMI-Lab/SparseKD/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":244640022,"owners_count":20485966,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-07-31T03:01:52.593Z","updated_at":"2025-03-20T15:31:18.836Z","avatar_url":"https://github.com/CVMI-Lab.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"# Towards Efficient 3D Object Detection with Knowledge Distillation\n\n---\n[Jihan Yang](https://jihanyang.github.io/), [Shaoshuai Shi](https://shishaoshuai.com/), [Runyu Ding](https://github.com/Dingry), [Zhe Wang](https://wang-zhe.me/), [Xiaojuan Qi](https://xjqi.github.io/)\n\nThis repository contains the official implementation of [Towards Efficient 3D Object Detection with Knowledge Distillation](https://openreview.net/pdf?id=1tnVNogPUz9), NeurIPS 2022\n\n\n## Changelog\n[2022-09-] Code release for Waymo results.\n\n## Introduction\nOur code is based on [OpenPCDet v0.5.2](https://github.com/open-mmlab/OpenPCDet/tree/v0.5.2).\nMore updates on OpenPCDet are supposed to be compatible with our code.\n\n## Model Zoo\n### Waymo Open Dataset\n#### Basic Models without KD\nSimilar to OpenPCDet, all models in the following table are trained with a single frame of 20% data (~32k frames) of all \nthe training samples on 8 GTX 1080Ti GPUs. \nNote that the validation are also carried on the 20% validation set with `DATA_CONFIG.SAMPLED_INTERVAL.test 5` \n(similar performance with 100% validation set).\n\n|                             Model (20% data)                              | LEVEL2 mAPH | Flops (G) | Acts (M) | Latency (ms) |\n|:-------------------------------------------------------------------------:|:-----------:|:---------:|:--------:|:------------:|\n|        [CP-Voxel](tools/cfgs/waymo_models/cp-voxel/cp-voxel.yaml)         |    64.29    |   114.8   |  101.9   |    125.70    |\n|      [CP-Voxel-S](tools/cfgs/waymo_models/cp-voxel/cp-voxel-s.yaml)       |    62.23    |   47.8    |   65.7   |    97.99     |\n|     [CP-Voxel-XS](tools/cfgs/waymo_models/cp-voxel/cp-voxel-xs.yaml)      |    61.14    |   36.9    |   58.4   |    88.19     |\n|    [CP-Voxel-XXS](tools/cfgs/waymo_models/cp-voxel/cp-voxel-xxs.yaml)     |    56.26    |   12.0    |   33.1   |    70.44     |\n|       [CP-Pillar](tools/cfgs/waymo_models/cp-pillar/cp-pillar.yaml)       |    59.09    |   333.9   |  303.0   |    157.90    |\n|  [CP-Pillar-v0.4](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.4.yaml)  |    57.55    |   212.9   |  197.7   |    103.37    |\n| [CP-Pillar-v0.48](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.48.yaml) |    56.27    |   149.4   |  142.3   |    81.87     |\n| [CP-Pillar-v0.64](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.64.yaml) |    52.81    |   85.1    |   88.0   |    54.52     |\n\n\n#### Sparse Distillation Models \n|                                       Model (20% data)                                        | LEVEL2 mAPH | Gains |\n|:---------------------------------------------------------------------------------------------:|:-----------:|:-----:|\n|      [CP-Voxel-S + SparseKD](tools/cfgs/waymo_models/cp-voxel/cp-voxel-s_sparsekd.yaml)       |    64.25    | +2.0  |\n|     [CP-Voxel-XS + SparseKD](tools/cfgs/waymo_models/cp-voxel/cp-voxel-xs_sparsekd.yaml)      |    63.53    | +2.4  |\n|    [CP-Voxel-XXS + SparseKD](tools/cfgs/waymo_models/cp-voxel/cp-voxel-xxs_sparsekd.yaml)     |    59.28    | +3.0  |\n|  [CP-Pillar-v0.4 + SparseKD](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.4_sparsekd.yaml)  |    59.24    | +1.7  |\n| [CP-Pillar-v0.48 + SparseKD](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.48_sparsekd.yaml) |    58.53    | +2.3  |\n| [CP-Pillar-v0.64 + SparseKD](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.64_sparsekd.yaml) |    55.82    | +3.0  |\n\n\nHere we also provide the performance of several models trained on the full training set and validate on the \nfull validation set.\n\n|                                       Model (100% data)                                       | LEVEL2 mAPH | \n|:---------------------------------------------------------------------------------------------:|:-----------:|\n|                  [CP-Voxel](tools/cfgs/waymo_models/cp-voxel/cp-voxel.yaml)                   |    65.58    | \n|                 [CP-Pillar](tools/cfgs/waymo_models/cp-pillar/cp-pillar.yaml)                 |    61.56    |\n|               [PV-RCNN++](tools/cfgs/waymo_models/pv_rcnn_plusplus_resnet.yaml)               |    69.46    |\n|      [CP-Voxel-S + SparseKD](tools/cfgs/waymo_models/cp-voxel/cp-voxel-s_sparsekd.yaml)       |    65.75    |\n|     [CP-Voxel-XS + SparseKD](tools/cfgs/waymo_models/cp-voxel/cp-voxel-xs_sparsekd.yaml)      |    64.83    |\n|    [CP-Voxel-XXS + SparseKD](tools/cfgs/waymo_models/cp-voxel/cp-voxel-xxs_sparsekd.yaml)     |    60.93    |\n|  [CP-Pillar-v0.4 + SparseKD](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.4_sparsekd.yaml)  |    61.60    |\n| [CP-Pillar-v0.48 + SparseKD](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.48_sparsekd.yaml) |    60.95    |\n| [CP-Pillar-v0.64 + SparseKD](tools/cfgs/waymo_models/cp-pillar/cp-pillar-v0.64_sparsekd.yaml) |    58.89    |\n\n\n#### Cross stage distillation\n|                                     Model (20% data)                                      | LEVEL2 mAPH | Flops (G) | Acts (M) | Latency (ms) |\n|:-----------------------------------------------------------------------------------------:|:-----------:|:---------:|:--------:|:------------:|\n|             [PV-RCNN++](tools/cfgs/waymo_models/pv_rcnn_plusplus_resnet.yaml)             |    67.80    |   123.5   |  179.7   |    435.9     |\n|                [CP-Voxel](tools/cfgs/waymo_models/cp-voxel/cp-voxel.yaml)                 |    64.29    |   114.8   |  101.9   |    125.7     |\n| [CP-Voxel + SparseKd](tools/cfgs/waymo_models/cp-voxel/cp-voxel_sparsekd_crossstage.yaml) |    65.27    |   114.8   |  101.9   |    125.7     |\n\n\nWe could not publicly provide the above pretrained models due to [Waymo Dataset License Agreement](https://waymo.com/open/terms/).\nTo access these pretrained models, please email us your name, institute, a screenshot of the Waymo \ndataset registration confirmation mail, and your intended usage. Please send a second email if we don't get back to you \nin two days. Please note that Waymo open dataset is under strict non-commercial license, so we are not allowed to share \nthe model with you if it will use for any profit-oriented activities.\n\n\n### Latency on Different Hardware\n![latency](./docs/latency.png)\n\n\nTo facilitate further researches, we provide latency measured in milliseconds as follows:\n\n|      Model      | 1060 + Spconv2.x | 1080Ti + Spconv2.x | A100 + Spconv2.x | \n|:---------------:|:----------------:|:------------------:|:----------------:|\n|     SECOND      |      84.56       |       50.65        |      40.44       | \n|   PointPillar   |      129.12      |       68.61        |      44.14       | \n|    CP-Voxel     |      125.70      |       74.25        |      56.13       | \n|   CP-Voxel-S    |      97.99       |       65.68        |      39.40       |\n|   CP-Voxel-XS   |      88.19       |       61.41        |      36.65       | \n|  CP-Voxel-XXS   |      70.44       |       53.86        |      38.01       |\n|    CP-Pillar    |      157.90      |       77.29        |      51.94       | \n| CP-Pillar-v0.4  |      103.37      |       56.32        |      57.07       |\n| CP-Pillar-v0.48 |      81.87       |       49.64        |      27.44       | \n| CP-Pillar-v0.64 |      54.52       |       36.23        |      19.56       |\n\n\n## Installation\n\nPlease refer to [INSTALL.md](docs/INSTALL.md) for the installation of `OpenPCDet`.\n\n\n## Getting Started\n\nPlease refer to [GETTING_STARTED.md](docs/GETTING_STARTED.md) to learn more usage about this project.\n\n\n## License\n`Our code` is released under the [Apache 2.0 license](LICENSE).\n\n## Acknowledgement\nOur code is heavily based on [OpenPCDet](https://github.com/open-mmlab/OpenPCDet). \nThanks OpenPCDet Development Team for their awesome codebase.\n\n\n## Citation \nIf you find this project useful in your research, please consider cite:\n\n```\n@inproceedings{yang2022towards,\n    title={Towards Efficient 3D Object Detection with Knowledge Distillation},\n    author={Yang, Jihan and Shi, Shaoshuai and Ding, Runyu and Wang, Zhe and Qi, Xiaojuan},\n    booktitle={Advances in Neural Information Processing Systems},\n    year={2022}\n}\n```\n```\n@misc{openpcdet2020,\n    title={OpenPCDet: An Open-source Toolbox for 3D Object Detection from Point Clouds},\n    author={OpenPCDet Development Team},\n    howpublished = {\\url{https://github.com/open-mmlab/OpenPCDet}},\n    year={2020}\n}\n```\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FCVMI-Lab%2FSparseKD","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FCVMI-Lab%2FSparseKD","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FCVMI-Lab%2FSparseKD/lists"}