raymon-tian / keras_realtime_multi-person_pose_estimation

the keras implement of the Realtime Multi-Person Pose Estimation

keras_Realtime_Multi-Person_Pose_Estimation

the keras implement of the Realtime Multi-Person Pose Estimation

keras_realtime_multi-person_pose_estimation's Issues

Did the authour use the mask_miss and the mask_all when generate the heatmap and vecmap?

could you share the lmdb file？

Could you share your trained weights?

Hi,
Could you share your trained weights?
Thanks

How long a runtime should take? (CPU)

Im using CPU i7 gen 6, 8gb dd4, ssd 256MB, Windows 7 with Tensorflow compiled with avx2 (without MKL)
for a single image (712x674) it takes over 16 seconds, where the prediction itself (without loading the model or reading the image) takes about 14 seconds.

is it a way to imporve this runtime issue without using GPU?

請問問什麽使用relu作爲激活函數

在L2就是lmdb的部分，每兩個點作爲部位連結的一個單位向量，應該是有正負數值的，而relu不能輸出負數，所以請問是什麽原因使用relu呢？

Why your heatmap is different from the office version?

Hi,
I have check the official version. Your detected heatmap is different from official version, but your final result is same with it. I have checked the codes, but I have not found the reason.
Could you interpret why they are different for me ?
Thanks

Questions about demo.ipynb

In the demo.ipynb, you shown the heatmap[:,:,2], it looks a little noisy, but the final result looks good.
Is the final result based on the your trained model and up-mentioned noisy heatmap??
Thanks very much.