search

Hzzone / pytorch-openpose

pytorch implementation of openpose including Hand and Body Pose Estimation.
2.07k stars 395 forks source link

Could you tell me the model used evaluation result on COCO2017? #19

Open Tomwmg opened 4 years ago

Tomwmg commented 4 years ago

This code is very nice. Thank you for your contribution. I just used your demo to test on coco2017 dataset, and I got the following results: Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.291 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.505 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.282 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.143 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.498 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.344 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.530 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.348 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.155 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.604 It seems so low. Could you tell me the mAP you test, or some test details? Thanks a lot.

Hzzone commented 4 years ago

https://github.com/Hzzone/pytorch-openpose/blob/f6dda44ab6c7e6c7537b6a5bb540eb119bc09a60/python/body.py#L23 I have adopted a low search scale to accelerate the running speed which may solve your problem.

Tomwmg commented 4 years ago

Thanks for your response. I found your setting 'scale_search = [0.5, 1.0, 1.5, 2.0] ', and I choose this setting, the result is following: Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.566 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.789 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.610 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.518 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.648 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.606 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.805 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.647 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.527 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.719 It seems lower than the match paper, of course, there must some test details I cant get. Does it work normally?

Hzzone commented 4 years ago

sorry, I have no idea about it.