I need advice!

I study Computer Science and my final year project is in 2 months from now and it consists of two parts:

1- a program to detect human body parts sizes ( neck, chest, leg, arm…) from a served image similar to this. ( Status: incomplete )

2- develop a Restful API to to serve clients where one provides one or two images and gets measurements back(1) . ( Status : Done)

My problem is part 1, I’m new to AI and I have been taking lessons in deep learning from fast.ai and thought of two ways to solve this problem:

1.1 I wanted to know if I can build a model that will learn from human photos(with different keypoints of chest and shoulders, etc… maybe ! )+ specified measurements dataset which is impossible to find !

1.2 Or should I take a different path and use computer vision with open cv and image processing like this link

I feel lost from reading complex research papers which I find hard to implement and I don’t think dl is the go to solution for this problem so any pieces of advice are welcomed !
P.S: I posted in this forum hoping to get advice because it is somehow related to fastai .

Its true, reading papers can often be very time consuming and difficult. Your better off reading an article. Here’s one thats done what you want using javascript. It looks like deep learning is definitely going to be a good way to approach this problem.

I think you could use SegmentationItemList with fastai to solve this problem. Similar to what is shown in lesson 3

When you say sizes do you mean like how long someones arm is?

1 Like

Thanks for this article I really appreciate it :grinning:, I’ll check it and yes by sizes I meant each person’s arm length,chest area length,waist,etc…

That’s going to be quite hard without depth perception?

I am thinking after body part segmentation of a person’s image I use a reference object which I know its dimensions and calculate the pixel per metric from it and apply it on each body part region maybe to get its size !
Object size measurements article

Hi. I am also doing my final project on a similar topic :sweat_smile:

I handle this problem like this.
Do segmentation and PoseEstimation, then compute number of pixels on the mask in the regions of shoulders, waist and hips and divide those numbers by the height. This way I get relative sizes of different parts.

2 Likes