|
| 1 | +# ACRN |
| 2 | + |
| 3 | +## 简介 |
| 4 | + |
| 5 | +<!-- [DATASET] --> |
| 6 | + |
| 7 | +```BibTeX |
| 8 | +@inproceedings{gu2018ava, |
| 9 | + title={Ava: A video dataset of spatio-temporally localized atomic visual actions}, |
| 10 | + author={Gu, Chunhui and Sun, Chen and Ross, David A and Vondrick, Carl and Pantofaru, Caroline and Li, Yeqing and Vijayanarasimhan, Sudheendra and Toderici, George and Ricco, Susanna and Sukthankar, Rahul and others}, |
| 11 | + booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition}, |
| 12 | + pages={6047--6056}, |
| 13 | + year={2018} |
| 14 | +} |
| 15 | +``` |
| 16 | + |
| 17 | +<!-- [ALGORITHM] --> |
| 18 | + |
| 19 | +```BibTeX |
| 20 | +@inproceedings{sun2018actor, |
| 21 | + title={Actor-centric relation network}, |
| 22 | + author={Sun, Chen and Shrivastava, Abhinav and Vondrick, Carl and Murphy, Kevin and Sukthankar, Rahul and Schmid, Cordelia}, |
| 23 | + booktitle={Proceedings of the European Conference on Computer Vision (ECCV)}, |
| 24 | + pages={318--334}, |
| 25 | + year={2018} |
| 26 | +} |
| 27 | +``` |
| 28 | + |
| 29 | +## 模型库 |
| 30 | + |
| 31 | +### AVA2.1 |
| 32 | + |
| 33 | +| 配置文件 | 模态 | 预训练 | 主干网络 | 输入 | GPU 数量 | mAP | log | json | ckpt | |
| 34 | +| :----------------------------------------------------------: | :------: | :----------: | :------: | :---: | :--: | :--: | :----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: | |
| 35 | +| [slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb](/configs/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb.py) | RGB | Kinetics-400 | ResNet50 | 32x2 | 8 | 27.1 | [log](https://download.openmmlab.com/mmaction/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb.log) | [json](https://download.openmmlab.com/mmaction/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb.json) | [ckpt](https://download.openmmlab.com/mmaction/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava_rgb-49b07bf2.pth) | |
| 36 | + |
| 37 | +### AVA2.2 |
| 38 | + |
| 39 | +| 配置文件 | 模态 | 预训练 | 主干网络 | 输入 | GPU 数量 | mAP | log | json | ckpt | |
| 40 | +| :----------------------------------------------------------: | :------: | :----------: | :------: | :---: | :--: | :--: | :----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: | |
| 41 | +| [slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb](/configs/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb.py) | RGB | Kinetics-400 | ResNet50 | 32x2 | 8 | 27.8 | [log](https://download.openmmlab.com/mmaction/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb.log) | [json](https://download.openmmlab.com/mmaction/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb.json) | [ckpt](https://download.openmmlab.com/mmaction/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb-2be32625.pth) | |
| 42 | + |
| 43 | +- 注: |
| 44 | + |
| 45 | +1. 这里的 **GPU 数量** 指的是得到模型权重文件对应的 GPU 个数。默认地,MMAction2 所提供的配置文件对应使用 8 块 GPU 进行训练的情况。 |
| 46 | + 依据 [线性缩放规则](https://arxiv.org/abs/1706.02677),当用户使用不同数量的 GPU 或者每块 GPU 处理不同视频个数时,需要根据批大小等比例地调节学习率。 |
| 47 | + 如,lr=0.01 对应 4 GPUs x 2 video/gpu,以及 lr=0.08 对应 16 GPUs x 4 video/gpu。 |
| 48 | + |
| 49 | +对于数据集准备的细节,用户可参考 [数据准备](/docs_zh_CN/data_preparation.md)。 |
| 50 | + |
| 51 | +## 如何训练 |
| 52 | + |
| 53 | +用户可以使用以下指令进行模型训练。 |
| 54 | + |
| 55 | +```shell |
| 56 | +python tools/train.py ${CONFIG_FILE} [optional arguments] |
| 57 | +``` |
| 58 | + |
| 59 | +例如:在 AVA 数据集上训练 ACRN 辅以 SlowFast 主干网络,并定期验证。 |
| 60 | + |
| 61 | +```shell |
| 62 | +python tools/train.py configs/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb.py --validate |
| 63 | +``` |
| 64 | + |
| 65 | +更多训练细节,可参考 [基础教程](/docs_zh_CN/getting_started.md#训练配置) 中的 **训练配置** 部分。 |
| 66 | + |
| 67 | +## 如何测试 |
| 68 | + |
| 69 | +用户可以使用以下指令进行模型测试。 |
| 70 | + |
| 71 | +```shell |
| 72 | +python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} [optional arguments] |
| 73 | +``` |
| 74 | + |
| 75 | +例如:在 AVA 上测试 ACRN 辅以 SlowFast 主干网络,并将结果存为 csv 文件。 |
| 76 | + |
| 77 | +```shell |
| 78 | +python tools/test.py configs/detection/acrn/slowfast_acrn_kinetics_pretrained_r50_8x8x1_cosine_10e_ava22_rgb.py checkpoints/SOME_CHECKPOINT.pth --eval mAP --out results.csv |
| 79 | +``` |
| 80 | + |
| 81 | +更多测试细节,可参考 [基础教程](/docs_zh_CN/getting_started.md#测试某个数据集) 中的 **测试某个数据集** 部分。 |
0 commit comments