ScaleNet

KO | EN

gitlite — search

#react #typescript #llm #android #machine-learning #rust #linux #macos #cli #kubernetes #vue #pytorch #security #mcp

(★ 54)

Data-Driven Neuron Allocation for Scale Aggregation Networks

ScaleNet 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# ScaleNet

By Yi Li, Zhanghui Kuang, Yimin Chen, Wayne Zhang

SenseTime.

### Table of Contents
0. [Introduction](#introduction)
0. [Citation](#citation)
0. [Approach](#approach)
0. [Trained models](#trained-models)
0. [Experiments](#experiments)
0. [GPU time](#gpu-time)

### Introduction

This is a PyTorch implementation of [Data-Driven Neuron Allocation for Scale Aggregation Networks](https://arxiv.org/pdf/1904.09460.pdf).(CVPR2019) with pretrained models.

### Citation

If you use these models in your research, please cite:

@inproceedings{Li2019ScaleNet,
        title={Data-Driven Neuron Allocation for Scale Aggregation Networks},
        author={Li, Yi and Kuang, Zhanghui and Chen, Yimin and Zhang, Wayne},
        booktitle={CVPR},
        year={2019}
	}

### Approach
<div align="center">
  <img src="https://github.com/Eli-YiLi/ScaleNet/blob/master/figures/architecture.png" width="300">
</div>
<p align="center">
  Figure 1: architecture of ScaleNet-50.
</p>

<div align="center">
  <img src="https://github.com/Eli-YiLi/ScaleNet/blob/master/figures/sablock.png" width="200">
</div>
<p align="center">
  Figure 2: scale aggregation block.
</p>

### Trained models
| Model | Top-1 err. | Top-5 err. |
|:-:|:-:|:-:|
| ScaleNet-50-light | 22.80 | 6.57 |
| ScaleNet-50 | 22.02 | 6.05 |
| ScaleNet-101 | 20.82 | 5.42 |
| ScaleNet-152 | 20.06 | 5.18 |

Pytorch:

```
from pytorch.scalenet import *
```
```
model = scalenet50(structure_path='structures/scalenet50.json', ckpt=None) # train from stratch
```
```
model = scalenet50(structure_path='structures/scalenet50.json', ckpt='weights/scalenet50.pth') # load pretrained model
```

The weights are available on [BaiduYun](https://pan.baidu.com/s/1NOjFWzkAVmMNkZh6jIcMzA) with extract code: f1c5

Unlike the paper, we used better training settings: increase the epochs to 120 and replace multi-step learning rate by cosine learning rate.

### Experiments

<div align="center">
  <img src="https://github.com/Eli-YiLi/ScaleNet/blob/master/figures/imagenet.png" width="300">
</div>
<p align="center">
  Figure 3: experiments on imagenet classification.
</p>

<div align="center">
  <img src="https://github.com/Eli-YiLi/ScaleNet/blob/master/figures/coco.png" width="300">
</div>
<p align="center">
  Figure 4: experiments on ms-coco detection.
</p>

### GPU time
| Model | Top-1 err. | FLOPs(10^9) | GPU time(ms)|
|:-:|:-:|:-:|:-:|
| ResNet-50 | 24.02 | 4.1 | 95 |
| SE-ResNet-50 | 23.29 | 4.1 | 98 |
| ResNeXt-50 | 22.2 | 4.2 | 147 |
| ScaleNet-50 | 22.2 | 3.8 | 93 |

TensorFlow:
(empty models of ResNet, SE-ResNet, ResNeXt, ScaleNet for speed test)
```
python3 tensorflow/test_speed.py scale|res|se|next
```

All networks were tested using Tensorflow with GTX 1060 GPU and i7 CPU at batch size 16 and image side 224 on 1000 runs.

Some static-graph frameworks like Tensorflow and TensorRT execute multi-branch models in parallel, while Pytorch and Caffe do not. So we suggest to deploy ScaleNets on Tensorflow and TensorRT.

ScaleNet

ScaleNet 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ScaleNet

ScaleNet 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README