It’s time for another INPUT session.
INPUT.3:
November 23, 2015, 7-9pm, 2 S Park St (SF)
RSVP requested
The format will be more rigid than last time.
• 7:00pm - Introductions
• 7:15pm - Daniel Kondermann, CTO of a German Computer Vision startup, will give a 30-minute talk, Crowd Truth (abstract below)
• 7:55pm - I will give a five minute demo of Gentle, a very-very-soon-to-be-released open source forced aligner (precisely aligns audio with transcripts)
• 8:00pm - Dinner will be provided, conversation is encouraged (but not required), and a CDG lab tour can be organized for those who haven’t seen it yet
Crowd Truth - Training Data Generation For Computer Vision
Pallas Ludens GmbH solves "unsolvable" computer vision tasks at scale with humans in the loop. We deliver accurate results for low and
high-level computer vision such as object tracks, labels, segmentations and even optical flow. Our services are regularly used by automotive,
medical and media companies such as Daimler and Zeiss who seek to test and train their methods for automated computer vision.
In this talk, I first introduce our business model which is based on inserting what we call nanotasks into computer games. The second part is
about example projects we did this year and focuses on the challenges with scene labeling for the Cityscapes Dataset, which was a very difficult
task even for trained annotators. Finally, I present a new ground truth dataset for urban autonomous driving which contains a fusion of measured
as well as manually annotated data.
You are welcome to invite others who might be interested, but please keep me informed on attendance so I can send entrance instructions and order the right amount of food.
Your correspondent,
R.M.O.