English | 简体中文
Now FastDeploy supports the deployment of the following models
Export the PP-TinyPose and PP-PicoDet models. Please refer to Model Export
Attention: The exported inference model contains three files, including model.pdmodel、model.pdiparams and infer_cfg.yml. FastDeploy will get the pre-processing information for inference from yaml files.
For developers' testing, part of the PP-PicoDet + PP-TinyPose（Pipeline）exported models are provided below. Developers can download and use them directly.
| Application Scenario | Model | Parameter File Size | AP(Service Data set) | AP(COCO Val Single/Multi-person) | Single/Multi-person Inference Time (FP32) | Single/Multi-person Inference Time（FP16) |
|---|---|---|---|---|---|---|
| Single-person Model Configuration | PicoDet-S-Lcnet-Pedestrian-192x192 + PP-TinyPose-128x96 | 4.6MB + 5.3MB | 86.2% | 52.8% | 12.90ms | 9.61ms |
| Multi-person Model Configuration | PicoDet-S-Lcnet-Pedestrian-320x320 + PP-TinyPose-256x192 | 4.6M + 5.3MB | 85.7% | 49.9% | 47.63ms | 34.62ms |
Note
- The accuracy of the keypoint detection model is based on the detection frame obtained by the pedestrian detection model.
- The flip operation is removed from the accuracy test with the detection confidence threshold of 0.5.
- The speed test environment is qualcomm snapdragon 865 with 4-thread inference under arm8.
- The Pipeline speed covers the preprocessing, inference, and post-processing of the model.
- In the accuracy test, images with more than 6 people (excluding 6 people) were removed from the multi-person data for fair comparison.
For more information: refer to PP-TinyPose official document