Nov 9, 2017
Qualcomm products mentioned within this post are offered by Qualcomm Technologies, Inc. and/or its subsidiaries.
The only thing a developer community likes more than a cool, new programming frontier (like artificial intelligence) is the chance to win cool, new hardware (like the Qualcomm Snapdragon 835 Mobile Hardware Development Kit) on which to develop it.
That’s why we launched the Qualcomm Artificial Intelligence Developer Contest in September. It’s our way of kick-starting and building up the community for AI developers around our Qualcomm Snapdragon Neural Processing Engine (NPE) SDK.
About 200 developers submitted an entry during Round 1 of the contest to take on-device machine learning and artificial intelligence to the next level on a Snapdragon mobile platform. The 20 entries shown below will now move to Round 2 of the contest, and our judges have awarded each of them a Snapdragon 835 Mobile HDK on which to bring their AI idea to life.
In other words, now the real fun starts.
AI: Absolutely Inspiring
The AI ideas we received covered a lot of ground and a wide variety of themes.
About one-third of entries approved by our judges fall into the category of fun, like photography, music, poetry and fashion. Several others are related to health, biometrics, diagnostics and early detection. Long-tail entries fall into IoT, assistive, agriculture, drones, learning and automotive. We’ve been amazed by the diversity of applications for training and using neural networks.
I think that the international breadth of entries has a lot to do with that diversity. Developers around the globe look at situations differently and come up with different ways to apply AI. About two-thirds of contest entries came from three countries - India, USA and Great Britain - with the long tail reaching as far as Turkey, Albania, Kenya, Honduras and Ghana.
And The Finalists Are...
Here, then, are the 20 finalists and a short summary of the AI applications they propose:
1. Francois Lemarchand - Great Britain - This mobile app automatically evaluates the photographs you take based on aesthetic criteria using a deep neural network (DNN).
2. Slava Vasiliev - Russia - Traditional market scales will be improved with 3D camera and machine language processing to identify the foods weighed and estimate the cost and nutritional value for each item.
3. Ashish Mokalkar - India - DeepAgro captures real-time images of wide-area farmlands using drones and allows farmers to easily identify crop diseases. The app predicts the ideal height of the crop, recommends possible causes of the disease and suggests adequate fertilizer usage for higher crop yield.
4. Bassel Ebeed - Egypt - This app will describe a scene captured by the camera in a mobile device or stored on the internet (Twitter, Facebook, etc.) and answer speech-based questions related to that scene. It will help visually impaired people, especially when they’re outdoors or browsing the web.
5. Pallab Sarkar - India - NeuralSense seeks to overcome the lack of a globally standardized sign language through an assistive application that empowers deaf and mute people to communicate efficiently among themselves and with people who hear normally.
6. Henry Ruiz Guzmán - Colombia - This freely available mobile app will allow scientists, researchers and farmers to extract physiological information from the images of plants (rice, beans and cassava to start), collected during the crop monitoring process.
7. Deane Landreth - New Zealand - To reduce the number of inevitable traffic accidents in drive-on-the-left-side countries, this application will monitor the road ahead and warn if it detects a vehicle on the wrong side of the road.
8. Phong Nguyen - Viet Nam - This is an autonomous drone that uses an integrated processor for accurate, real-time landings in areas where GPS is unreliable or absent.
9. Hakan Özkelemci - Turkey - This mobile app will employ deep learning techniques to guide users with step-by-step interactive tutorials and instant feedback for teaching them to play an instrument such as the guitar.
10. Eric Kotonya - Kenya - AIRRAPP is a poetic chat mobile app that lets anyone chat with an AI friend in a personalized, entertaining, context-sensitive way through rhyming sentences generated by the AI engine.
11. Gabriel Grand - United States - To forecast impending seizures minutes or even hours before they occur, this app runs a recurrent neural network (RNN) on a patient’s mobile device, monitors biometrics through wearable devices and performs computationally intensive time series analysis.
12. Beven Sandengu - Great Britain - Coin-It is an augmented reality app that uses coins as markers to make precise distance/area measurements of objects in real time, as a convenient tape measure for mobile devices.
13. Ognjen Todic - United States - This will be an SDK that provides on-device speech recognition for Snapdragon-powered devices.
14. Md. Khairul Alam - Bangladesh - To help farmers take perfect care of their plants, this mobile app will give information on plant diseases by processing images of plant leaves taken through the device’s camera.
15. Francis Walugembe - Uganda - This artificial intelligence algorithm and software product has the potential to detect the sound of a baby crying, analyze the sound and decode its meaning for parents or caregivers.
16. Navin Bhaskar - India - With machine learning and AI, skin cancer can be detected at early stage, so this app will help people in rural areas where it is difficult to obtain a medical exam.
17. Sunny Aditya - India - Emma is an app that will help people in figuring out which clothes look good on them.
18. Bipul Das - Canada - Using the front camera of the device, this app will capture facial image streams and apply on-device AI to monitor users’ underlying emotional state and provide recommendations when they engage in social networking or similar media apps.
19. Sergey Konvisarov - Finland - From a random drawing, SNPE-accelerated AI will recognize lines, curves, ovals, polylines, relative displacement, size and other attributes to control a module that generates in real time different styles of popular music.
20. Devang Mohan - India - This app can apply style transfer filters to a real-time camera feed, giving users the artistic freedom to choose a frame that goes with how they feel during photo composition, instead of having to pick a filter that goes with a photo they have already taken.
In Round 2, our 20 finalists will have their chance to develop using the Snapdragon NPE SDK, our software development kit designed for tuning the performance of AI applications relying on Caffe/Caffe2 and TensorFlow deep learning frameworks.
Shout-out to the judges
Speaking of Caffe2, which is Facebook’s open source deep learning framework, Qualcomm Technologies and Facebook are collaborating to optimize Caffe2 for the Snapdragon NPE SDK. As part of that collaboration, joining our judges from Qualcomm Technologies for the final round of the contest is Yangqing Jia, a research scientist manager currently leading Facebook's AI platform team. His team develops general-purpose AI solutions that serve as the backbone of Facebook AI products, such as ranking, computer vision, natural language processing, speech recognition and mobile AI and AR.
Yangqing has a track record of developing open source deep learning frameworks, and is known as the creator/co-creator of Caffe, TensorFlow, Caffe2 and ONNX over the years. Before Facebook, Yangqing was a research scientist at Google Brain, after obtaining his PhD at UC Berkeley in Computer Science.
There’s no place like Barcelona in late February, which is why the Grand Prize winner of the Artificial Intelligence Developer Contest will win his or her choice of a trip for two to Mobile World Congress 2018 in Barcelona, Spain, or a lump-sum, cash payment of US$10,000.
I hope the variety of AI innovations this contest has inspired will inspire you as well. If you’re ready to get your AI game on, download the Snapdragon NPE SDK and have a look through the reference app we’ve included to get you started quickly. Also, install Caffe2 and TensorFlow on your development machine to build and train neural network models that will work with the SDK.