czr

they flipped a switch (this one, under "clear to go" subheading https://youtu.be/ToDMceo0aDs?t=20). tesla planning still mostly is done by C/C++ code.

nuro https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...

> Yet another throw-away product from a self-proclaimed environmentally friendly company.

$79 for airpods max battery service [0]. user-replaceable battery certainly would be nicer. but to describe as "throw-away" seems extreme.

[0] https://support.apple.com/airpods/repair/service#battery

iirc they project small green region around the actors and real props, so that ambient light and reflections are still mostly correct but they can also get clean matte out.

how do you differentiate open-mindedness from a closed-minded preference in other direction (i.e. straight men who specifically want to out-earn their spouses)?

important to remember that header image is result of professional concept artist drawing over gaugan result (bottom left). explanation of paper is not awful but not great, would recommend to read the original https://arxiv.org/pdf/1903.07291.pdf

actual paper https://aip.scitation.org/doi/10.1063/1.5129958

video 1 (response measurement): https://aip.scitation.org/doi/suppl/10.1063/1.5129958/suppl_...

video 2 (picking up block): https://aip.scitation.org/doi/suppl/10.1063/1.5129958/suppl_...

video 3 (hexapod): https://aip.scitation.org/doi/suppl/10.1063/1.5129958/suppl_...

video 4 (human climbing tiled wall): https://aip.scitation.org/doi/suppl/10.1063/1.5129958/suppl_...

video 5 (human climbing concrete wall): https://aip.scitation.org/doi/suppl/10.1063/1.5129958/suppl_...

typo: under supported podcasts, heavyweight is listed twice.

for those interested, there's a longer discussion of this idea at https://jameshfisher.com/2014/05/11/your-syntax-highlighter-... (2014)

hack to switch to light mode, if you don't want to do it systemwide (paste into console):

    Array.from(document.styleSheets).forEach(ss => Array.from(ss.rules).forEach((ssr,i) => {if (ssr.cssText && ssr.cssText.includes("dark")) {ssr.parentStyleSheet.deleteRule(i)}}))

he hasn't worked on YOLO, only NLP. YOLO is another example of a well-known, successful researcher (Joseph Redmon) writing an informal paper

for those who aren't familiar with the author, he previously worked at metamind / salesforce research doing nlp and has published many successful nlp papers [0]. he opted to write an informal paper for this project (similar to yolov3 [1]), but the work itself should still be taken seriously.

[0] https://scholar.google.com/citations?user=AolIi4QAAAAJ

[1] https://pjreddie.com/media/files/papers/YOLOv3.pdf

haven't read paper yet, but this is a silly demo. "turn 10% of pixels black" is not a good baseline, should use nearest-neighbor interp (or something) to fill holes in the "sparse" video for fair comparison. also, you can clearly see in hd video that it's temporally unstable ("shimmering"), which is the same problem nvidia has with dlss since forever; need to build temporal smoothing in or users will hate it.

problem is most likely that your initialization are bad (see e.g. https://openreview.net/pdf?id=H1gsz30cKX for explanation). make sure to use variance scaling, taking activation into account (relu cuts variance by half). probably you need to multiply all initial kernel weights by 2. make sure that initial prediction of model before training is same order of magnitude as typical target, not saturated to zero or other extreme value. batchnorm and skip connections can also ease problems of bad initialization, so worth trying.

for specifics, would recommend to simply read the code. for overall context, geohot has done interviews ([0], [1]) and that give overview of comma/openpilot at high level.

[0] http://www.autonocast.com/blog/2018/7/13/91-hotz-takes-with-... [1] https://youtu.be/iwcYp-XT7UI?t=1605

good start op. a good next step is add in temporal context (previous frame information) so model may resolve ambiguous cases. see e.g. breakdown of comma.ai openpilot model here (from their twitter, [0]), or the karpathy talk as well.

also, when presenting results, prefer to include a longer demo or sub-demos that show strengths and failure cases, and move to the top of the post rather than bottom imo. for a given reader, implementation details are either confusing and uninteresting (if not subject-matter expert) or predictable and uninteresting (if an expert, you've seen many similar before); audience for implementation details is very small number of people who want to sit down and replicate or check your work. but demos / analysis is always novel and interesting to anyone, so lead with that :)

[0] https://medium.com/@chengyao.shen/decoding-comma-ai-openpilo...

for those interested in the different ml methods attempted for the challenge, many of the top kagglers in this contest posted rundowns of their solution [0].

for those interested in kuronet [1] and the competition alex gave a presentation here (en, [2]) as did tarin (ja, [3]).

for those interested in classical japanese literature, you can see the tools they've built here [4]. tarin is also very active on twitter [5].

[0] https://www.kaggle.com/c/kuzushiji-recognition/discussion

[1] https://arxiv.org/pdf/1910.09433.pdf

[2] https://youtu.be/e2-mVBvygm8?t=1428

[3] https://youtu.be/e2-mVBvygm8?t=537

[4] https://mp.ex.nii.ac.jp/kuronet/

[5] https://twitter.com/tkasasagi

they made an extended abstract for ismir: http://archives.ismir.net/ismir2019/latebreaking/000036.pdf

methodology is a separate u-net per instrument type to predict a soft mask in spectrogram space (time x frequency), then they apply that mask to the input audio. fairly standard.

melodyne's editing interface lets you remove different notes from a polyphonic track. so if it's just vocals + other tonal sounds, you can manually remove the other tonal sounds. example: https://youtu.be/2ZjdDatxTaQ?t=83

messed around with the 2stem model for a bit and it's reasonably good. I think phonicmind is still a bit better - phonicmind tends to err on the side of keeping too much, while the 2stem model tries to isolate aggressively and often damages the vocal as a result (distorting words by losing some harmonics, or losing quiet words entirely)

example:

https://files.catbox.moe/wjruiv.mp3 (phonicmind)

https://files.catbox.moe/uuzot3.mp3 (spleeter 2stem)

you can hear spleeter does better at actually taking out the bass drums, but phonicmind never loses or distorts any part of the vocal, while 2stem occasionally sounds like singing is through metal tube (harmonics are missing). will try to read instructions more carefully and see if there's some way to fix.

timestamped link: https://overcast.fm/+R7DXB8Hbs/14:08

Agreed. I think this is an instance of:

> It is only by harsh experience that we learn which principles take priority over which other principles; as mere words they all sound equally persuasive.

ah, looks like admins fixed the title. this comment is in relation to the title the article uses: "All the lonely people ... are men", which is unsupported by the data in the self-same article.

> all

18% of men, 12% of women; 18/30 is... 60%? huh?

yup! best of luck, love your research.

if you haven't already seen it, https://github.com/dbolya/yolact generates very sharp instance masks very quickly, may be of interest

does anyone have more information on this? only reference I can find is that blog post; I can't find the specific section of the linked specs that discusses such a utility

waifu2x still much nicer for art of course (comparison: https://i.imgur.com/4QkIUOc.png 2x, https://i.imgur.com/pQDuIpl.png 4x)

but for the stated purpose this looks pretty good. for example, 720p [https://giant.gfycat.com/AccomplishedBelatedBlueshark.webm] to 1440p [https://giant.gfycat.com/FluidBlissfulCob.webm] test. is subtle, improves video, and runs fine (tested via mpv, https://mpv.io/manual/master/#options-glsl-shaders).