Tailor Master

Project Title

Tailor Master

Project Area of Specialization Artificial IntelligenceProject Summary

Tailoring is one of the most common and low investment business in Pakistan. We are planning to make an application that will take a small 15 second video of a customer standing in a hand’s up position. The person has to rotate around 360 degree slowly starting from the right and come back to his/her position and bring his arms down and straight within 15 seconds. The user will put the phone on a flat surface and make sure his whole body comes in the screen. After making the video, the application will ask about the height and gender of the person. The height will be used to get reference points to extract measurements. And gender will be used to recommend designs. The model will be trained to classify all the measurements of a person using the video. The customer can choose from some predefined options of shalwar and kameez and some simple designs. On the other hand, the customer will be given a drawing platform where he/she can draw his customized design for the dress on an outline of a dummy human or enter an image of a dress too. The customer can enter manual details as well if he/she wants. At the end the app will generate a report of the measurements and the design the customer wants which they can email, WhatsApp or send through any source to the tailor. Furthermore, the system will be artificially intelligent to recommend user styles from other users’ entered styles or default styles to a customer based on the previously used templates that are already saved.

Project Objectives

Our goal is to develop an app that takes a short clip of user. We will train a model that will use the clip to extract measurements of shalwar and kameez. The App will have GUI to select different designs and view designs of other users on the basis of gender. People can make customized designs too. The application will generate a report that will contain the current measurements of the user and the design they want. And after a couple of uses the application will recommend the user designs according to their previous preferences.

Project Implementation Method

Design methodology of front end of the application (Iteration 1): The following use cases are implemented: • Recording video front end of the application • Entering measurements manually. • Allowing upload of image of dress design. • Viewing report option but at this time only manually entered measurements are shown along with uploaded image. • We have given the connection to screen of drawing customized dress design the module is incomplete right now.

Data collection (Iteration 2): ImageNet is an image dataset organized according to the WordNet hierarchy. Each meaningful concept in WordNet, possibly described by multiple words or word phrases, is called a "synonym set" or "synset". There are more than 100,000 synsets in WordNet, majority of them are nouns (80,000+). In ImageNet, we aim to provide on average 1000 images to illustrate each synset. Images of each concept are qualitycontrolled and human-annotated. In its completion, we hope ImageNet will offer tens of millions of cleanly sorted images for most of the concepts in the WordNet hierarchy. On 30 September 2012, a convolutional neural network (CNN) called AlexNet achieved a top-5 error of 15.3% in the ImageNet 2012.

Design and Training Model for Measurement extraction (Iteration 2): The image frames are extracted from the video at first. As in the next iteration we need to integrate the models and in the last iteration deploy them over the application. We had to work in TensorFlow-lite or pyTorch-lite. We are using a variation of MobileNet model that is used to detect the pose of a person and difference between human joints. We trained it over the ImageNet dataset’s 37 person class. We modified the OpenCV DNN Example to use the Tensorflow MobileNet Model, instead of Caffe Model from CMU OpenPose. The original openpose.py from OpenCV example only uses Caffe Model which is more than 200MB while the Mobilenet only 7MB. MobileNet is a simple but efficient and not very computationally intensive convolutional neural networks for mobile vision applications. MobileNet is widely used in many real-world applications which includes object detection, fine-grained classifications, face attributes, and localization. Moreover, we used the bodypix to extract different body parts of the body. The implementation of mobileNet is shown in mobilenet.py file and the variation of how further I used it for body measurement extraction in shown in the colab notebook. We actually tried to integrate the functionality of OpenPose Caffe model on the mobileNet features so that we could do the required work in less space.

Benefits of the Project

An application that helps the user to get measurements without using a measurement scale and then communicating one by one everything to tailor. The application will generate a report that will contain measurements and customized design of the user that can be shared. The application can give recommendations too.

Technical Details of Final Deliverable

•Extracting 3D Measurements from a 2D clip

•Artificially Intelligent model to make recommendations

•Deploying the models on the android phone

•Integration of models and the GUI

Final Deliverable of the Project Software SystemCore Industry ITOther Industries Agriculture Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	35000
Android Mobile Phone	Equipment	1	35000	35000