<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Tanishasingha on Medium]]></title>
        <description><![CDATA[Stories by Tanishasingha on Medium]]></description>
        <link>https://medium.com/@tanishasingha37657?source=rss-8e6d5b623935------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*mNfZwkE57jgryJt6</url>
            <title>Stories by Tanishasingha on Medium</title>
            <link>https://medium.com/@tanishasingha37657?source=rss-8e6d5b623935------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 23:16:35 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tanishasingha37657/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[From Body Measurements to a 3D Digital Body: A Virtual Try-On Prototype]]></title>
            <link>https://medium.com/@tanishasingha37657/from-body-measurements-to-a-3d-digital-body-89372e24e910?source=rss-8e6d5b623935------2</link>
            <guid isPermaLink="false">https://medium.com/p/89372e24e910</guid>
            <category><![CDATA[fashion-technology]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[3d-modeling]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[computer-vision]]></category>
            <dc:creator><![CDATA[Tanishasingha]]></dc:creator>
            <pubDate>Mon, 07 Sep 2026 04:01:24 GMT</pubDate>
            <atom:updated>2026-09-07T04:18:00.038Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8ykvhWHjfAJksXDXHsJK8A.png" /></figure><p>Shopping for clothes online is not an easygoing experience for many of us, especially those who have trouble finding the right size of garment without trying it on. Different brands have a different fit for the same size label. A size ‘M’ of one brand may not be the same as the size ‘M’ of another brand.</p><p>As a frequent shopper of online clothing brands myself, I’ve always thought how less tedious this whole experience would be if only there was a digital-me that could try on the clothes before I press on that ‘Add to cart’ button. That’s what my project aims to solve. This Virtual Try-On prototype uses input data in the form of body measurements of chest, waist, hips, height and weight of the body, breadth of the shoulder, breadth of the hip, and circumference of the thigh to create a 3D body mesh using the SMPL (Skinned Multi-Person Linear) body model.</p><h4><strong>How the System Works</strong></h4><p>The system follows a measurement-to-mesh pipeline:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SfVSCcPBKMvi7hmARnp6yw.png" /><figcaption>Basic Workflow</figcaption></figure><p>The user enters a total of 8 body measurements and selects the type of body model: male, female or neutral. Once the user provides body measurements, they are sent as input to a Multi-Layer Perceptron (MLP) neural network, which outputs the predicted 10 shape parameters. The 10 shape parameters are called beta values.</p><p>Then, these beta values are used to create the shape of the body mesh using SMPL.</p><p>Instead of taking a common dataset to derive the body type, 3 different and specific datasets are created for the model, and separate MLPs are trained on these datasets.</p><h4><strong>Training the Models</strong></h4><p>Since real anthropometric datasets paired with 3D body scans are scarce, I generated synthetic training data instead: sampling a range of SMPL body shapes and extracting their corresponding measurements. This produced three datasets (synthetic_male.csv, synthetic_female.csv, synthetic_neutral.csv), each used to train its own network (mlp_male.pt, mlp_female.pt, mlp_neutral.pt). At inference time, the body type selected by the user determines which trained network and which SMPL template are used.</p><h4><strong>Results</strong></h4><p>The models were evaluated using:</p><p><strong>1.</strong> <strong>Mean Per-Vertex Error (MPVE)</strong></p><p>Measures the average distance between corresponding vertices of the predicted mesh and ground-truth mesh. The lower this value the better is the mesh prediction. MPVE of all three models is approximately 5 mm, which indicates that the predicted mesh is about 5 mm away from the corresponding vertices of the ground truth mesh.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/441/1*a6cDsaGwYf-v95O3dTbfoA.png" /></figure><p><strong>2.</strong> <strong>Beta MSE (Mean Squared Error)</strong></p><p>Measures how close the MLP’s predicted SMPL beta values are to the true beta values. Lower the value better the prediction.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/442/1*ZkIQd2dj2h_ricYvUY61tw.png" /></figure><p><strong>3.</strong> <strong>Per-Measurement Reconstruction Error</strong></p><p>Measures the difference between the body measurements extracted from the predicted mesh and those extracted from the ground-truth mesh.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/805/1*a3ehBnbS4PJCHc0egURsnw.png" /></figure><p>These results indicate that the MLP can learn a useful relationship between anthropometric measurements and SMPL body-shape parameters.</p><h4>Output Meshes</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/757/1*EwYDxKSTUbKvtx7FnhuAgg.png" /><figcaption>Male Body Mesh</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/758/1*e4ACFEAfmtCFrO0j2IFQzQ.png" /><figcaption>Female Body Mesh</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/754/1*bHrmGTynhIXl6zsC90VMhg.png" /><figcaption>Neutral body mesh</figcaption></figure><h4>Conclusion and What’s Next</h4><p>The prototype demonstrates a feasible solution for 3D body mesh generation from limited body measurements. It currently allows the user to input their measurements, select a type of model and obtain the resulting 3D body mesh.</p><p>The concept can be extended into a fully-functioning system for virtual clothing try-on by adding clothing meshes, garment fitting, body-cloth collision detection, and realistic rendering. The current prototype may improve through the use of a more representative, real data on anthropometric properties for better results.</p><p>Ultimately, the goal of this project is to eliminate uncertainty in online purchases and ensure that a customer is able to see what the clothing would look like on their individual virtual body.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=89372e24e910" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>