search

michalfaber / tensorflow_Realtime_Multi-Person_Pose_Estimation

Multi-Person Pose Estimation project for Tensorflow 2.0 with a small and fast model based on MobilenetV3
Other
216 stars 65 forks source link

Did you test coco MAP? #4

Closed zerollzeng closed 4 years ago

zerollzeng commented 4 years ago

Thanks for you great job! did you test map on coco dataset, I run your pretrained mobilenetv2 model on some sample iamges , and the result seems not very accurate. eg image

michalfaber commented 4 years ago

Despite quite good performance, heat maps based solution is not very accurate. This is because output heat maps have lower resolution than input image due to the downsampling steps in a deep neural network (x8) For input image of size 224x224, heat map resolution is only 28x28. This causes quantization errors.

lengyuner commented 4 years ago

Where do you find the pretrained mobilenetv2 model?