Computer vision (CV) is often touted as the autonomous vehicle technologies' force. However, CV promises prominence in many fields as corporate leaders transform their business models to be competitive by looking into new market prospects through innovation or partnership. 

Sensing the opportunity and the need, DataLyca was founded in 2021 as a startup building CV products. We came across the startup at the Digital India Week 2022 in Gandhinagar. 

The start

The startup is working on two CV products: BirdiIns (player tracker) in the MVP stage and Radiology AI in the tech evaluation stage.

"When we started working on the sports-related use case, we evaluated different options, such as using Inertial Movement Unit (IMU) sensors and other tracking devices to get player data. However, as none of the approaches rendered reliable and consistent data, as part of exploration, we built the PoC (Proof of Concept) using computer vision and finalized the tech stack," said Karunakar Reddy R, Founder and Director, DataLyca. 

"We see tremendous opportunities in using computer vision across the domains. Though we struggled to build a model during the initial days, now we have two experts dedicatedly working on computer vision-related tasks," he added. 

With over 15 years of experience, Karunakar's early stint with MindTree and Thomson Reuters as a tech leader provided the foundation. He earlier co-founded a startup called ElewaSoftwares and had successfully built a product on VR/AR.  

Products to offer

Though the company has built a framework for analyzing CT scan data, the current and sole focus now is on BirdieIns (Birdie means shuttle cock and Ins for insights). Birdies is a Badminton AI coach that helps players to understand the areas they are performing well and areas they have to focus more on. BiridieIns is available both in the B2B and B2C formats.

• In the B2B space, BirdieIns provides comprehensive data along with auto-generated feedback. BirdieIns also has a provision for the coach to annotate video frames or any graph before sharing feedback data in PDF format. In this busy world, everyone wants to realize insights quickly; BirdieIns help to truncate and get a professional look by improvising captured video as well as truncating all breaks. 

One of the academies currently uses the product as part of the stabilization phase. "We have achieved more than 95% in service detection, 94% in shot detection, and 98% in player detection/speed," said Karunakar.

• B2C product is targeted at any player. It has an option for players to capture video and upload using the mobile app to view all insights. Currently, this solution is patent-pending, and we plan to launch the MVP of this product by September 2022.

On dataset and training

"As my son is a badminton player and I own a sports club, we did not face many difficulties collecting videos. It took longer than expected to get a good accuracy model," said Karunakar. 

The startup used data from five venues and also introduced noise in images randomly to train the model; this helps the model to be more robust in detecting objects even if frames are not of good quality. Karunakar further says that when he explained the product and the list of features, he was intrigued to see some parents and coaches were very adaptive in allowing us to record the video. Parents show great interest as they can see the valuable insights about their kids' performance. He still remembers one occasion when many of the parents called me multiple times to check when they would be able to take a look at insights in the mobile app. 

This was long back when full automation was not implemented. Currently, the product is fully automated from video recording, uploading as well processing of video data, and viewing insights in the mobile app.

Challenges at hand

Certainly, challenges were there, and we have faced a lot of issues with regards to,

• Issues in terms of building capabilities. Initially, I hired a consultant who helped us build the core part of the model and algorithms. Even now, we fall back on him for blockers. 
• When it comes to data, although I initially got access to record videos in two of the clubs, it was a bit difficult to convince other coaches when nothing was there to show what we were into. 
• Further, there was difficulty in deciding on the camera and the lens, and we faced many difficulties in procuring them. 

Despite these odds, Karunakar believes, "The harder the risks, the more potential for the product."

Talking about the privacy issues, Karunakar said, "We make sure that the data or video of the player is safe and that only the coach or player can view the data. We are also dividing data based on geographics and two verticals of the product( for B2B and B2C) and are committed to ensuring a safe data protection ecosystem."

Future Plans

"We are currently focused on releasing BirdieIns to a larger audience in India before getting into overseas markets. We want to be a one-stop shop for any coaching-related activities for Badminton games. Our plan in sports includes implementing a similar solution for other racquet games such as Table tennis, Tennis, and Squash," said Karunakar.

He also believes that government can do much more, especially promoting AI startups. They can help to get the required data for startups to train models and funding and can help startups to tap overseas opportunities. When we started a startup in 2012, there was hardly any scheme aimed at promoting startups. I am fortunate to have good support from the STPI IMAGE COE program in terms of mentoring and connecting us to the concerned team/person to get all the required helps. He believes that many a thing needs to be done to boost the startup ecosystem as a whole.