Computer Vision Meetup Roundup #4 2017
Welcome to the recap of the Computer Vision Meetup April Edition! We were very proud that the meetup was hosted in a special location: at the top of the IBM building in Vienna, with an amazing view over the city!
VISUAL RECOGNITION WITH IBM WATSON
Manuel Fuchs, Cognitive Computing Consultant started his talk by showing a video of IBM, how people perceive it and how they actually want to present themselves nowadays. He pointed out that IBM actually is pioneer in intelligent and cognitive systems and that the Watson network is just growing constantly. Generally, Watson has all of their services online as an API service on the Watson Developer Cloud. You can go there and try it out for free.
After the general intro on IBM, Florian Rosenberg, Computer Vision of IBM Watson specialist took over.
Florian started by explaining that Visual Recognition actually brings structure to images, recognizing what’s displayed in them.
He then explained the terms of general tagging, custom classification, similarity search and face detection.
The general tagging generally gives an answer to the question “What in the world is in this image?”. The custom training and classification is the ability to train your own classifiers for your images. Florian pointed out that custom classification is something that none of IBM’s competitors can offer at the moment.
Florian then talked about how to measure the accuracy of the Visual Recognition. Watson validates the general tagging model by A/B testing it with users. They can therefore choose which set of tags they prefer. You can find all the details of the presentation in the linked file above!
Decomposing videos into units and building their relationships by PIK’D
“pik’d uses machine learning to process accelerometer and gyroscope data. The solution offers unmatched speed, accuracy and energy efﬁciency in highlight detection and classification.”
Every GoPro user knows the pain of going through all the video material you just shot and finding that one great moment. Pik’d is a startup that is here to solve this problem. They use machine learning to automatically classify the content of sports videos. When you’ve made a video while snowboarding and you had a great jump, Pik’d is going to find that jump and label it, without you having to go through all your video material!
Uploading huge video files to a cloud, where they can be further processed is a pain. That’s why they came up with an app solution for smartphones and for the laptop. For this solution they use motion sensors which are included in all POV Cameras. This means that besides recording the video itself, they camera also saves a motion sensor data via gyroscope sensor. When you connect your camera to your smartphone or laptop, the data gets further processed and labelled and you don’t have to go through hours of video material. Pik’d categorizes and sorts out the exciting video material right on the smartphone. This means you just connect the GoPro Camera to the phone and can watch your greatest moments there and share them in an instant! Of course your raw video material is also still there. The newest GoPros send the video and gyroscope data to the smartphone via bluetooth, where they can then be processed and labaled by the Pik’d app.
To train their Deep Neuronal Networks, they used the video data of hundreds of athletes around the world.
By the way: Pik’d is hiring! Go check out their site, and recommend them if you know anyone in the Computer Vision scene!
HOLD A TALK YOURSELF!
You have a project or topic you would like to talk about or you know someone, who would like to share his/her experiences and knowledge at our Computer Vision Meetup? Please contact us!
It is great to see how our community is growing each month, so if you haven’t already – don’t forget to join our meetup group ! ?