Speaker 0:  for decades, filming at millions of frames per second required a $100,000 camera that only elite research labs could ever afford. but a massive breakthrough called the drum camera just slashed the price by tagged by 90% using exact same micromirror chip found in standard school classroom projectors. can canadian researchers and meta have built an ultra fast camera hitting 4,800,000 frames per second using cheap off the shelf commercial components. this allows an imaging system that typically costs 6 figures to be construction constructed for under $10,000. this completely democratizes advanced scientific imaging and paves the way for hyper precise, low cost lidar sensors in everyday autonomous vehicles. what do you think about this, john?

Speaker 1:  i like this. this is cool. i love that it's canadian too. so let's talk about the problem and how drum changes it. so traditional high speed cameras, they have a brutal engineering challenge. right? so in order to capture millions of frames per second, the sensors themselves must read out enormous amounts of data pretty much instantaneously, and that requires specialized cmos sensors. it requires high bandwidth memory and sophisticated electronics and massive storage systems. so the faster you wanna record, the more expensive the camera becomes. this is why a lot of these scientific grade cameras from manufacturers like phantom, photron, and specialized imaging can cost well over a $100,000. but there's also a trade off between frame rate, resolution, sensitivity, recording duration. so, traditionally, increasing one often then sacrifices another, and this is one of the biggest bottlenecks in high speed imaging. so the genius of drum is that it doesn't primarily rely on a faster sensor, which is super cool. instead, it uses diffraction optics to encode temporal information before the light reaches the camera sensor. so different moments of a fast event are mapped into different spatial locations through a diffraction based gating system. and so the camera captures all this information in a single exposure, and computational reconstruction separates the encoded information into individual frames. so i i'm hesitant on this because we're this is part of a broader trend called computational imaging. so we're getting into reconstruction and things like that, but it's cool because reduce, it does reduce the cost. so rather than solving problems purely with hardware, what researchers are doing and what they did here is they're looking to combine optics, signal processing, computational reconstruction, and machine learning. so, the result is a performance that would traditionally cost you a $100,000. what do you think about this? so what do you think about computational imaging?

Speaker 0:  think of this update like trying to capture seven rapid fire photos of a speeding bullet using a single camera. instead of trying to force a physical shutter to mechanically open and close millions of times a second, which breaks the hardware, you place a grid of tiny hyper fast pivoting mirrors right in front of the camera sensor. as the light from the camera fast moving event rushes in, these cameras instantly tilt mirrors instantly tilt in a preprogrammed sequence acting like optical trapdoors that route each microsecond slice of the action to a different corner of the sensor all at once. so, yeah, i think it's cool.

Speaker 1:  what do you think about relying more on software reconstruction than actual hardware? do you think we're getting a pure image? or like, my concern is that once we go to software, we run into potential malicious players when it comes to this kind of stuff. right? so

Speaker 0:  it's too late for that, isn't it?

Speaker 1:  it is late. yeah.

Speaker 0:  like, i mean, isn't i every phone camera you is mostly software now?

Speaker 1:  pretty much. yeah. you're right.

Speaker 0:  yeah. so, i think we're beyond that. so people are gonna be malicious. people are gonna be and we just have to, you know, accept a new world. right? people, use did you hear it?

Speaker 1:  we might have covered this with samsung. you know what they did with one of their galaxy s phones is when you took a picture of the moon, actually it used it reconstructed a really clear picture of the moon, so it wasn't the actual picture you took.

Speaker 0:  i did not see that. but, yeah, this there you go. i mean yeah. and people say there's two. most of the, pictures of the earth you've seen is not, an actual photo. they're digital reconstructions.

Speaker 1:  yeah.

Speaker 0:  so, you know, is this techno technology dangerous in, a fraud or sure. but look at ai image generation and video generation now. do we even need these $10,000 cameras to create stuff that people are you know? so people

Speaker 1:  are making movies on their iphones. right? so you guys might have heard craig at the start saying meta was involved in this. this. right? so why meta? why was meta involved? so meta has invested heavily in advanced computer vision, augmented reality, as we know, virtual reality, and machine perception technologies. so, ultra fast imaging, it has potential in arvr sensing, spatial computing, motion tracking, next generation lidar, like you said, autonomous systems. so this collaboration, it definitely has commercial implications for future sensing technologies. you combine this with the color lidar, you've got a crazy system.

Speaker 0:  absolutely. and to validate the camera could accurately measure real world physic dynamics physical dynamics, researchers didn't just fire lasers. they literally used a system to film the chaotic microsecond level explosion of bubbles inside an ordinary carbonated soft drinks to map fluid microdynamics. so the drum camera turns a 100,000 luxury research tool into a $10,000 commodity, proving that clever engineering with mass produced components will always disrupt the overpriced legacy hardware. now that this tech is poised to radically drop the price of autonomous vehicle sensors, would you just trust a self driving car more knowing it tracks the road 4,800,000 frames per second, or are you still keeping your hands firmly on the wheel? let us know in the chat below.