HRNet.md 4.5 KB
Newer Older
W
WuHaobo 已提交
1 2 3
# HRNet系列

## 概述
littletomatodonkey's avatar
littletomatodonkey 已提交
4
HRNet是2019年由微软亚洲研究院提出的一种全新的神经网络,不同于以往的卷积神经网络,该网络在网络深层仍然可以保持高分辨率,因此预测的关键点热图更准确,在空间上也更精确。此外,该网络在对分辨率敏感的其他视觉任务中,如检测、分割等,表现尤为优异。
littletomatodonkey's avatar
littletomatodonkey 已提交
5

littletomatodonkey's avatar
littletomatodonkey 已提交
6
该系列模型的FLOPS、参数量以及FP32预测耗时如下图所示。
littletomatodonkey's avatar
littletomatodonkey 已提交
7

littletomatodonkey's avatar
littletomatodonkey 已提交
8
![](../../images/models/T4_benchmark/t4.fp32.bs4.HRNet.flops.png)
littletomatodonkey's avatar
littletomatodonkey 已提交
9

littletomatodonkey's avatar
littletomatodonkey 已提交
10 11 12
![](../../images/models/T4_benchmark/t4.fp32.bs4.HRNet.params.png)

![](../../images/models/T4_benchmark/t4.fp32.bs4.HRNet.png)
littletomatodonkey's avatar
littletomatodonkey 已提交
13

littletomatodonkey's avatar
littletomatodonkey 已提交
14
目前PaddleClas开源的这类模型的预训练模型一共有7个,其指标如图所示,其中HRNet_W48_C指标精度异常的原因可能是因为网络训练的正常波动。
W
WuHaobo 已提交
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29


## 精度、FLOPS和参数量

| Models      | Top1   | Top5   | Reference<br>top1 | Reference<br>top5 | FLOPS<br>(G) | Parameters<br>(M) |
|:--:|:--:|:--:|:--:|:--:|:--:|:--:|
| HRNet_W18_C | 0.769  | 0.934  | 0.768             | 0.934             | 4.140        | 21.290            |
| HRNet_W30_C | 0.780  | 0.940  | 0.782             | 0.942             | 16.230       | 37.710            |
| HRNet_W32_C | 0.783  | 0.942  | 0.785             | 0.942             | 17.860       | 41.230            |
| HRNet_W40_C | 0.788  | 0.945  | 0.789             | 0.945             | 25.410       | 57.550            |
| HRNet_W44_C | 0.790  | 0.945  | 0.789             | 0.944             | 29.790       | 67.060            |
| HRNet_W48_C | 0.790  | 0.944  | 0.793             | 0.945             | 34.580       | 77.470            |
| HRNet_W64_C | 0.793  | 0.946  | 0.795             | 0.946             | 57.830       | 128.060           |


littletomatodonkey's avatar
littletomatodonkey 已提交
30
## 基于V100 GPU的预测速度
W
WuHaobo 已提交
31

littletomatodonkey's avatar
littletomatodonkey 已提交
32
| Models      | Crop Size | Resize Short Size | FP32<br>Batch Size=1<br>(ms) |
littletomatodonkey's avatar
littletomatodonkey 已提交
33 34
|-------------|-----------|-------------------|--------------------------|
| HRNet_W18_C | 224       | 256               | 7.368                    |
littletomatodonkey's avatar
littletomatodonkey 已提交
35 36
| HRNet_W30_C | 224       | 256               | 9.402                    |
| HRNet_W32_C | 224       | 256               | 9.467                    |
littletomatodonkey's avatar
littletomatodonkey 已提交
37 38 39 40
| HRNet_W40_C | 224       | 256               | 10.739                   |
| HRNet_W44_C | 224       | 256               | 11.497                   |
| HRNet_W48_C | 224       | 256               | 12.165                   |
| HRNet_W64_C | 224       | 256               | 15.003                   |
littletomatodonkey's avatar
littletomatodonkey 已提交
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55




## 基于T4 GPU的预测速度

| Models      | Crop Size | Resize Short Size | FP16<br>batch_size=1<br>(ms) | FP16<br>batch_size=4<br>(ms) | FP16<br>batch_size=8<br>(ms) | FP32<br>batch_size=1<br>(ms) | FP32<br>batch_size=4<br>(ms) | FP32<br>batch_size=8<br>(ms) |
|-------------|-----------|-------------------|------------------------------|------------------------------|------------------------------|------------------------------|------------------------------|------------------------------|
| HRNet_W18_C | 224       | 256               | 6.79093                      | 11.50986                     | 17.67244                     | 7.40636                      | 13.29752                     | 23.33445                     |
| HRNet_W30_C | 224       | 256               | 8.98077                      | 14.08082                     | 21.23527                     | 9.57594                      | 17.35485                     | 32.6933                      |
| HRNet_W32_C | 224       | 256               | 8.82415                      | 14.21462                     | 21.19804                     | 9.49807                      | 17.72921                     | 32.96305                     |
| HRNet_W40_C | 224       | 256               | 11.4229                      | 19.1595                      | 30.47984                     | 12.12202                     | 25.68184                     | 48.90623                     |
| HRNet_W44_C | 224       | 256               | 12.25778                     | 22.75456                     | 32.61275                     | 13.19858                     | 32.25202                     | 59.09871                     |
| HRNet_W48_C | 224       | 256               | 12.65015                     | 23.12886                     | 33.37859                     | 13.70761                     | 34.43572                     | 63.01219                     |
| HRNet_W64_C | 224       | 256               | 15.10428                     | 27.68901                     | 40.4198                      | 17.57527                     | 47.9533                      | 97.11228                     |