# Comparision between Light HR Net and Crowd Pose

**Crowd Pose is a based on Alpha Pose**

**Various COCO Dataset Evaluation metrics described [here](http://cocodataset.org/#detection-eval)**

**Some CrowdPose Dataset Metrics:**
APHard = CrowdIndex of (0.8 to 1)
APMedium = CrowdIndex of (0.1 to 0.8)
APEasy = CrowdIndex of (0 to 0.1)

| *Parameter of Comparison* | [Light HR Net](https://arxiv.org/pdf/1911.10346.pdf) (depends on backbone) | [AlphaPose](https://arxiv.org/abs/1612.00137) | [CrowdPose](https://arxiv.org/pdf/1812.00324.pdf) |
| ------------------------- | ------------ | --- | ---------- |
| | | | |
| Number of Parameters |**2.9M to 7.4M**|28.1M|equal to or more than 28.1M|
| FLOPs |**1.0G to 1.8G**|26.7G|more than 26.7G|
| COCO AP | 68.7 to 70.4 |**72.3**| |
| COCO AP50 |90.2 to **91** |89.2| |
| COCO AP75 |76.9 to **78.9**|**79.1**| |
| COCO APM |65.9 to **67.7**|**68.0**| |
| COCO APL |74.3 to 76 |**78.6**| |
| COCO mAP @0.5:0.95 | |70.1| **70.9** |
| COCO mAR @0.5:0.95 | |74.4| **76.4** |
| CrowdPose Dataset mAP @0.5:0.95| |61.0| **66.0** |
| CrowdPose Dataset mAP @0.5| |81.3| **84.2** |
|CrowdPose Dataset mAP @0.75| |66.0| **71.5** |
|CrowdPose Dataset mAR @0.5:0.95| |67.6| **72.7** |
|CrowdPose Dataset mAR @0.5 | |86.7| **89.5** |
|CrowdPose Dataset mAR @0.75| |71.8| **77.5** |
|CrowdPose Dataset APEasy | |71.2| **75.5** |
|CrowdPose Dataset APMedium | |61.4| **66.3** |
|CrowdPose Dataset APHard | |51.1| **57.4** |
|CrowdPose Dataset FPS* | |**10.9**| 10.1 |
|COCO Dataset FPS** | **17 to 11** | | |

**This Result Was Obtained using two Nvidia 1080Ti GPUs. But the newer version on Github promises better speed and accuracy*
***This Result Was Obtained with just one Intel i7-8700K CPU
machine*