This is my final version for Project Three. My original idea started as a continuation of Project Two as I liked being able to use bears as I found trying to make the features too human made my sketches look very weird. I am happy with how my project turned out as I think the bears still look cute but I was still able to communicate the parameters of the faces through the use of position data and the training data.
I have used a combination of position landmarks and using the values on my sliders and training the images to get the features to show up. I used the positions to mark the eyes nose, mouth and then ears (to the eyebrows) as I wanted the expression of the person in the photo to still be clear.
The values controlled by my sliders are:
Tilt value Face Width Face Colour - 5 different face tones which affects the base face colour and detail colour Hair length - of the little hairs on top of the head Hair Colour - 6 different colours that affect the inside of the ears and the hairs on top of the head Eye Colour - four colours, dark brown, light brown, blue, green Blush - either has blushed cheeks to distinguish between male and female - females have blush cheeks males do not.
My sample images don't have a relation to bears as I found it was very hard to find an efficient sample image that has a direct correlation. Instead I chose photos from TV shows that I really enjoy - specifically ones that have a lot of diversity and inclusion among their cast as I think that is important for the pictures used in this assignment. I am happy with how my parameters are reacting with the faces on these images after doing the training. I think it does a good job on picking up on gender, hair colour and skin tone and I think that the positioning variables help to capture the expression of the person in the photo too.
In the nearest neighbours setting I am quite happy with how my training has affected the outcome of this. I can see the similarities and I think it particularly picks up on angle and expression similarities so I am glad that I included some expression from the position parameters. In the training quiz I think I can usually pick out which one it should be which is good. However I do think there might be something a little buggy with the first training image as sometimes when I reload it changes its values even though I have saved them and entered them into training_values.
Overall I am happy with how my final project turned out as I found this assignment a little daunting at the start but I think I managed to communicate the facial features whilst maintaining the bear style from my Assignment Two.