So this week I have been attempting to figure out how to match my poses from an image already uploaded. Currently, my approach is getting all the x,y positions of the keypoints (lefteye etc) and placing these into an array. I am also doing the same for the poses. To try and match these poses I am trying an approach that was inspired by Google's move mirror in which they try to match poses with GIFs.

To do this, one matching method they used is cosine similarity. This measure of similarity is between two vectors: basically, it measures the angle between them and returns -1 if they’re exactly opposite, 1 if they’re exactly the same. Importantly, it’s a measure of orientation and not magnitude.

Currently, I am still struggling with getting this to work due to the fact that the package I'm using to compute the similarity is having a bit of trouble to determine. Further down the track I will continue to work on this

To reflect on this week's progress, I probably should have tried this cosine similarity method earlier as I had found it earlier but just brushed it off because I thought it was too difficult. I realised that this was wrong and I should have probably tried to implement this earlier and if I got it incorrect I could have more time to try and get it right. Furthermore, I could have also asked my team members about what they have found so then if there was an easier method, I could try to implement it. To rectify these changes, I will try to make myself more available to team members and I will try to try new things earlier.