So I'm not a morning person. As a new parent of a 1-year-old, I realized I can get an extra 45 minutes of sleep in the morning by leaving books in our baby's crib for her to look through when she wakes up. One night I left Computer Engineering for Babies in there, and after waking the next morning I looked on the baby monitor and I saw her turning the pages of the book, methodically pressing both buttons for the AND gate, one button at a time for the OR gate, etc. It was probably one of my proudest father moments so far. She was about 18 months at that point, and I hadn't seen her really understand the gates before.
Commercial products in the space have been around for several years but can only screen for a small number of genetic diseases at a time. For example, Natera's Spectrum screens for chromosomal abnormalities (eg Down Syndrome) plus one custom genetic problem for a single condition of greatest concern. https://www.natera.com/womens-health/spectrum-preimplantatio...
Some polygenic screening services that look at a much larger number of conditions are just reaching market now (eg Orchid Health), though our ability to map genes correlated with polygenic traits to phenotype results is still in its early stages. https://www.orchidhealth.com/
Matterport's camera uses a mixture of structured light, deep learning, and other techniques to build a 3D point cloud from each capture. Then, the app aligns the different captures together to produce a 3D mesh.
It's also possible to use consumer 360 cameras with Matterport's app, though accuracy is lower. In this case, the 3D-from-2D estimation primarily happens using deep learning.
FWIW, I don't think he's creating a monocrop forest -- he's planting a range of species that form an interdependent web.
This is also being done in locations that were forests a few hundred years ago before humans cut them down. The mature forest is a self-sustaining ecosystem, but it needs extra water for a few decades until it reaches that point. https://www.westhawaiitoday.com/2020/01/12/hawaii-news/yisha...
The world has a lot of land, though. Before humans came along, there were roughly 6 trillion trees in the world. Now there are about 3 trillion, mainly due to forests being cut down for agriculture, wood, etc. Getting back the other 3 trillion trees would sequester an additional 75 gigatons of CO2 per year using your figure of 25kg/tree/year. Human activity emits 45gt/year according to his article.
I think the actual costs will be a lot lower as this will take many years of experimentation before scaling, and by then the cost of the solar panels (which I think is the vast majority of the cost) will likely have continued to ride its exponential downward cost curve. One source: https://earthtechling.com/solar-energy-costs-trends/
I agree, but I still think of this as a useful part of a complete solution. It's something that buys us time to shift away from fossil fuels.
With current tech, we could drastically reduce emissions by moving to a mix of nuclear (the modern, safe kind) for baseload plus wind and solar for peak, but it's going to take few decades to build it all out, and during the intervening period we have a lot of excess CO2 we need to sequester.
I'm not an expert in this area, but from what I can tell:
Young rapidly growing forests sequester carbon more quickly than mature forests, but mature forests still sequester some carbon, possibly into the soil. (One source https://www.sciencedaily.com/releases/2008/09/080910133934.h... )
Separately, the regrown forest could be maintained in a state of peak carbon sequestration by doing selective logging from time to time, and as a bonus it would be profitable to do so.
Some thoughts on finite CO2-absorbing capacity:
- If these forests are sustainably logged, we can take some of the carbon out and sequester it in buildings (or even just bury it), allowing the forest to regrow and absorb more carbon.
- I have a theory that presently discussed global climate change solution only needs to buy us a maximum of 100 years, and by then our tech for dealing with silly problems like excess carbon will have almost surely advanced beyond recognition. Given rates of progress in chemistry, physics, and materials science over the last two centuries, this isn't crazy.
Lots of things! Since 3D convolutional networks are very limited in their maximum resolution, most of the interesting things you can do involve learning on RGB+D images via a 2D CNN. A lot of tasks on images (segmentation, identification etc) are easier when you have even partial depth data as input.
I've done a lot of work in this area, and I can say that this is significantly faster and higher quality than the other Kinect-based 3D reconstruction techniques out there such as RGBDemo ( http://www.youtube.com/watch?v=Cldf7UdFq1k ).
It's also clear just how much of an advantage having a 3D sensor is for reconstruction when you compare this against 2D-camera-based 3D reconstruction software like Photofly.