jefft255

28 ms·

241.
by jefft255 8y ago
That’s not really what a tensor is; this simplification is due to tensorflow I think?
242.
by jefft255 8y ago
It’s nice indeed. Not special, but certainly smooth to use. Never used it outside of emacs though.
243.
by jefft255 8y ago
They are fundamentally different: to summarize, Waymo follows the HD maps approach when you have a very precise map of a given city, with both metric (actual 3D shape of the environment) and semantic (lanes, sidewalks, signs) information, i
244.
by jefft255 8y ago
Really cool, especially now that Autodesk's photogrammetry software is no longer available for free. I think this is a good example of academic/public research resulting in useful free software!
245.
Robot forest mapping for forest inventory(youtube.com)
2 points by jefft255 8y ago|0 comments
246.
by jefft255 8y ago
This is very much in use when possible. They are called fiducial markers, or sometimes registration targets, registration markers etc.
247.
by jefft255 8y ago
Well of course there is a limit, mapping at this scale absolutely requires GPS of loop-closure of some sort. Myself, working in forests, I mapped at most 100 meters by 100 meters but some people in my lab are going as large as 1 kilometer b
248.
by jefft255 8y ago
Hey, I'm familiar with your work! I'm currently submitting similar work using a Husky and a Velodyne HDL-32. I don't have the problem you mention with my sensor. See: https://www.youtube.com/watch?v=V-Q-XWSWT-
249.
by jefft255 8y ago
I (and OP for that matter) do mapping with sensors with accuracies that are around 2 cm. I don't know where you got that 1 cm requirement from. ICP/SLAM drift will happen even with a perfect sensor. It really depends on the scale
250.
by jefft255 8y ago
If you're mapping corn fields, GPS + IMU will yield very good results. I wouldn't use any kind of SLAM in a farm field, it will probably worsen the position given by the GPS + IMU!!
251.
by jefft255 8y ago
We can do that already by recording a rosbag and then reading the data back on a laptop. Or by port forwading, changing rosmaster and blablabla. This is just a quicker way to diagnose stuff. I do forest robotics, and when I'm in the wo
252.
by jefft255 8y ago
I am installing it on our lab's Clearpath Husky right now. Thank you!!
253.
by jefft255 8y ago
I don’t know about the rest of the world but in Quebec calculus is not taught with any epsilon-delta formalism. You don’t encounter that until mathematical analysis, which is in the curriculum of a very few majors in uni.
254.
by jefft255 8y ago
Not to be peculiar, but I don't know if approximating the hessian using the gradient counts as a second order method. I was talking about "full-blown" second order methods where you compute de hessian through AD. Furthermore,
255.
by jefft255 8y ago
Nope! Sorry I can’t take more time to explain but there is no second derivative used in Adam.
256.
by jefft255 8y ago
You known, ML researchers are aware of the enormous body of research in optimization. A lot of these methods simply do not work or scale well to deep neural networks. Second order methods have been tried and are not worth the cost.
257.
by jefft255 8y ago
The article seems good (although dated). See http://www.jmlr.org/papers/v18/17-468.html for an in depth survey from last year. But reading the title, I was like "well... sure?". Bad title for this articl
258.
by jefft255 8y ago
Lookup spherical CNNs and tensor field networks, both are rotation invariant using spherical harmonics. They are not really used in practice however.
259.
by jefft255 8y ago
Tree structural models: https://www.mdpi.com/2072-4292/5/2/491 Forest inventory: https://www.sciencedirect.com/science/article/pii/S092427161... And much more :) Source: I am d
260.
by jefft255 8y ago
I know, but for me it was on by default.
261.
by jefft255 8y ago
My new iphone will actually ask me to confirm that I am not driving before unlocking when it thinks I am
262.
by jefft255 8y ago
Why? I don't see any indication of that in the trailer. I just hope that it is a classical turn based pokémon game like I loved to play.
263.
by jefft255 8y ago
You're right! When talking about function approximation most theorems work on compact sets only (for most purposes that's a good enough hypothesis).
264.
by jefft255 8y ago
They are not "equivalent". Polynomials can only produce polynomials, but these polynomials can be made to be as close as you want to any continuous function (see Stone-Weierstrass theorem). For example, a NN with a ReLU activation
265.
by jefft255 8y ago
I second this. Switching from a Galaxy S7, the battery life for my Xr has been amazing!
266.
by jefft255 8y ago
a) this is not the same problem. The second equation you suggest could have a different minimum than the first, e.g. if f(x) goes to negative infinity but only when g(x) != 0. b) because this is a much more complex topic, for which you need
267.
by jefft255 8y ago
It depends, some C++11 features are widely used, like smart pointers. I agree that most of the features are seldom used and add tons bloat to the language.
268.
by jefft255 8y ago
I remember reading that he still uses a DOS computer to write his novel.
269.
by jefft255 8y ago
This could work, but it is important to note that a lot of ML algorithms trained in a closed domain (no "other" class) will be pretty bad at knowing what they don't know. This is an open problem in ML.
270.
by jefft255 8y ago
A pre-print is what it is: a version you put online before actually publishing your paper in a conference or a journal. In no field whatsoever an online pre-print is considered equivalent or sufficient compared to a peer-reviewed publicatio

More ›