HKS Pose Estimation — Day 2/30

Abdelrahman Helaly
2 min readJul 7, 2021

--

Practical Part

Today i will start with the practical part and will complete the remaining 2 apps

First of all, the TF implementation ran on my GPU and i am not the biggest fan of TF in general, so i decided to search for a pytorch implementation, i found this implementation.

App 3: Planking Pose Corrector

Very similar to the previous app built, Multiple distances and angles were selected from the given key points that are inferred.

Course Code

Head to left wrist distance (Used points: 0,7)

Head to right wrist distance (Used points: 0,4)

Angle between left shoulder and left wrist (Used points: 7,6,5)

Angle between right shoulder and right wrist (Used points: 4,3,2)

Angle between left hip and left ankle (Used points: 11,12,13)

Angle between right hip and right ankle (Used points: 8,9,10)

Code to perform the logic on the Angles and distances.

So Generally, All Practical application in this course follow the same pattern, detect key points, find the required angle and/or distance and perform if condition logic.

With that being said, I would rate the course 2/5 stars.

Overall, my practical experience with Open Pose wasn’t all pleasant, for instance mediapipe (Google’s) performs alot better on my current setup than openpose.

Theoretical Part of Open Pose

Today i read the first 5 pages of the total 14.

i will focus tomorrow on summarizing until i finish.

--

--