# RegNet
> [Designing Network Design Spaces](https://arxiv.org/abs/2003.13678)
<!-- [BACKBONE] -->
## Abstract
In this work, we present a new network design paradigm. Our goal is to help advance the understanding of network design and discover design principles that generalize across settings. Instead of focusing on designing individual network instances, we design network design spaces that parametrize populations of networks. The overall process is analogous to classic manual design of networks, but elevated to the design space level. Using our methodology we explore the structure aspect of network design and arrive at a low-dimensional design space consisting of simple, regular networks that we call RegNet. The core insight of the RegNet parametrization is surprisingly simple: widths and depths of good networks can be explained by a quantized linear function. We analyze the RegNet design space and arrive at interesting findings that do not match the current practice of network design. The RegNet design space provides simple and fast networks that work well across a wide range of flop regimes. Under comparable training settings and flops, the RegNet models outperform the popular EfficientNet models while being up to 5x faster on GPUs.
<div align=center>
<img src="https://user-images.githubusercontent.com/40661020/143971942-da50f719-61e9-43bd-9468-0dbfbe80284e.png"/>
</div>
## Introduction
We implement RegNetX and RegNetY models in detection systems and provide their first results on Mask R-CNN, Faster R-CNN and RetinaNet.
The pre-trained modles are converted from [model zoo of pycls](https://github.com/facebookresearch/pycls/blob/master/MODEL_ZOO.md).
## Usage
To use a regnet model, there are two steps to do:
1. Convert the model to ResNet-style supported by MMDetection
2. Modify backbone and neck in config accordingly
### Convert model
We already prepare models of FLOPs from 400M to 12G in our model zoo.
For more general usage, we also provide script `regnet2mmdet.py` in the tools directory to convert the key of models pretrained by [pycls](https://github.com/facebookresearch/pycls/) to
ResNet-style checkpoints used in MMDetection.
```bash
python -u tools/model_converters/regnet2mmdet.py ${PRETRAIN_PATH} ${STORE_PATH}
```
This script convert model from `PRETRAIN_PATH` and store the converted model in `STORE_PATH`.
### Modify config
The users can modify the config's `depth` of backbone and corresponding keys in `arch` according to the configs in the [pycls model zoo](https://github.com/facebookresearch/pycls/blob/master/MODEL_ZOO.md).
The parameter `in_channels` in FPN can be found in the Figure 15 & 16 of the paper (`wi` in the legend).
This directory already provides some configs with their performance, using RegNetX from 800MF to 12GF level.
For other pre-trained models or self-implemented regnet models, the users are responsible to check these parameters by themselves.
**Note**: Although Fig. 15 & 16 also provide `w0`, `wa`, `wm`, `group_w`, and `bot_mul` for `arch`, they are quantized thus inaccurate, using them sometimes produces different backbone that does not match the key in the pre-trained model.
## Results and Models
### Mask R-CNN
| Backbone | Style | Lr schd | Mem (GB) | Inf time (fps) | box AP | mask AP | Config | Download |
| :---------: | :-----: | :-----: | :------: | :------------: | :----: | :-----: | :------: | :--------: |
| [R-50-FPN](../mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py)| pytorch | 1x | 4.4 | 12.0 | 38.2 | 34.7 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_r50_fpn_1x_coco/mask_rcnn_r50_fpn_1x_coco_20200205-d4b0c5d6.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_r50_fpn_1x_coco/mask_rcnn_r50_fpn_1x_coco_20200205_050542.log.json) |
|[RegNetX-3.2GF-FPN](./mask_rcnn_regnetx-3.2GF_fpn_1x_coco.py)| pytorch | 1x |5.0 ||40.3|36.6|[config](https://github.com/open-mmlab/mmdetection/tree/master/configs/regnet/mask_rcnn_regnetx-3.2GF_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-3.2GF_fpn_1x_coco/mask_rcnn_regnetx-3.2GF_fpn_1x_coco_20200520_163141-2a9d1814.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-3.2GF_fpn_1x_coco/mask_rcnn_regnetx-3.2GF_fpn_1x_coco_20200520_163141.log.json) |
|[RegNetX-4.0GF-FPN](./mask_rcnn_regnetx-4GF_fpn_1x_coco.py)| pytorch | 1x |5.5||41.5|37.4|[config](https://github.com/open-mmlab/mmdetection/tree/master/configs/regnet/mask_rcnn_regnetx-4GF_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-4GF_fpn_1x_coco/mask_rcnn_regnetx-4GF_fpn_1x_coco_20200517_180217-32e9c92d.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-4GF_fpn_1x_coco/mask_rcnn_regnetx-4GF_fpn_1x_coco_20200517_180217.log.json) |
| [R-101-FPN](../mask_rcnn/mask_rcnn_r101_fpn_1x_coco.py)| pytorch | 1x | 6.4 | 10.3 | 40.0 | 36.1 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/mask_rcnn/mask_rcnn_r101_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_r101_fpn_1x_coco/mask_rcnn_r101_fpn_1x_coco_20200204-1efe0ed5.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_r101_fpn_1x_coco/mask_rcnn_r101_fpn_1x_coco_20200204_144809.log.json) |
|[RegNetX-6.4GF-FPN](./mask_rcnn_regnetx-6.4GF_fpn_1x_coco.py)| pytorch | 1x |6.1 ||41.0|37.1|[config](https://github.com/open-mmlab/mmdetection/tree/master/configs/regnet/mask_rcnn_regnetx-6.4GF_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-6.4GF_fpn_1x_coco/mask_rcnn_regnetx-6.4GF_fpn_1x_coco_20200517_180439-3a7aae83.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-6.4GF_fpn_1x_coco/mask_rcnn_regnetx-6.4GF_fpn_1x_coco_20200517_180439.log.json) |
| [X-101-32x4d-FPN](../mask_rcnn/mask_rcnn_x101_32x4d_fpn_1x_coco.py) | pytorch | 1x | 7.6 | 9.4 | 41.9 | 37.5 | [config](https://github.com/open-mmlab/mmdetection/tree/master/configs/mask_rcnn/mask_rcnn_x101_32x4d_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_x101_32x4d_fpn_1x_coco/mask_rcnn_x101_32x4d_fpn_1x_coco_20200205-478d0b67.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_x101_32x4d_fpn_1x_coco/mask_rcnn_x101_32x4d_fpn_1x_coco_20200205_034906.log.json) |
|[RegNetX-8.0GF-FPN](./mask_rcnn_regnetx-8GF_fpn_1x_coco.py)| pytorch | 1x |6.4 ||41.7|37.5|[config](https://github.com/open-mmlab/mmdetection/tree/master/configs/regnet/mask_rcnn_regnetx-8GF_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-8GF_fpn_1x_coco/mask_rcnn_regnetx-8GF_fpn_1x_coco_20200517_180515-09daa87e.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-8GF_fpn_1x_coco/mask_rcnn_regnetx-8GF_fpn_1x_coco_20200517_180515.log.json) |
|[RegNetX-12GF-FPN](./mask_rcnn_regnetx-12GF_fpn_1x_coco.py)| pytorch | 1x |7.4 ||42.2|38|[config](https://github.com/open-mmlab/mmdetection/tree/master/configs/regnet/mask_rcnn_regnetx-12GF_fpn_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-12GF_fpn_1x_coco/mask_rcnn_regnetx-12GF_fpn_1x_coco_20200517_180552-b538bd8b.pth) | [log](https://download.openmmlab.com/mmdetection/v2.0/regnet/mask_rcnn_regnetx-12GF_fpn_1x_coco/mask_rcnn_regnetx-12GF_fpn_1x_coco_20200517_180552.log.json) |
|[RegNetX-3.2GF-FPN-DCN-C3-C5](./mask_rcnn_regnetx-3.2GF_fpn_mdconv_c3-c5_1x_coco.py)| pytorch | 1x |5.0 ||40.3|36.6|[config](https://github.com/open-mmlab/mmdetection/tre
妄北y
- 粉丝: 2w+
- 资源: 1万+
最新资源
- 国际象棋检测7-YOLO(v5至v9)、COCO、CreateML、Darknet、Paligemma、TFRecord数据集合集.rar
- 【一文搞懂:什么是集成学习-原理+python代码】
- 德普微一级代理 DP3145D SOT23-6 USB PD 协议单口控制器
- java毕设项目之学习视频资源库的系统+jsp(完整前后端+说明文档+mysql).zip
- 基于java实现的快速排序代码
- java毕设项目之游泳会员管理系统+vue(完整前后端+说明文档+mysql+lw).zip
- 巨潮资讯网5000只股票orgId-dict加密字典
- java毕设项目之中文学习系统的设计与实现+jsp(完整前后端+说明文档+mysql+lw).zip
- 德普微一级代理 DP021N03FGLI DFN5*6 DPMOS N-MOSFET 30V 180A 1.8mΩ
- java毕设项目之zy+养老服务平台设计+vue(完整前后端+说明文档+mysql+lw).zip
- 【java毕业设计】SpringBoot+Vue幼儿园管理系统 源码+sql脚本+论文 完整版
- 德普微一级代理 DP100N06MGL PDFN3.3*3.3 TRMOS N-MOSFET 60V, 8mΩ, 45A
- mongoose-free-6.9
- dnSpy-net-win32-222.zip
- 国际象棋检测8-YOLO(v5至v9)、COCO、CreateML、Darknet、Paligemma数据集合集.rar
- cvtocc-shanghai.hbm
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈