As a note, for Japanese text deepl is widely used even by Japanese people. From eng to jpn it may not choose properly nuanced words though, but it largely produces acceptable translations.
In my experience, being a workplace or even academia, there's as many types of work environments as types of teachers.
In the end, management (or founders if it's a startup) set the tone and who they end up hiring (thus, who will manage your work). And management is just normal people in the end.
I guess because at server farm level, performance/efficiency translate to real million USD savings. In general, at scale ends (the cloud and the embedded) this matters a lot. In resource limited environments like raspberry pi, this design philosophy wins over many users between DIY and the private industry.
For training, more GPU RAM will allow you train with greater resolutions in less time and better performance.
Before feeding data to a model, it needs to be resized to a "network dimension" (YOLOv4 default is 416x416 px if I recall correctly). For training, it will group several samples and train with them at the same time, in "batches". For better generalization you want bigger batches (so more different images are feed at the same time).
With a 3060 (non-Ti) you'll have 12GB of GPU RAM, with that I think you can run the default settings (network size, batch size and subdivision of batches) for the YOLOv4 model. If you want to go to 512px, you might have to increase the subdivision (create more subbatches) or reduce the batch size.
If I recall correctly, you could find 3070 with less than 12GB of RAM, so in trying to purse faster training times (I'm not talking about inference, using the model to actually recognize something) you might not be able to train with a broader range of options that can improve your accuracy.
It would seem that it depends on the architecture you will be using. Whether is a ARM, GPU, mobile GPU processor, etc. This is comment from the author of YOLOv4 mentions that NanoDet is more suitable for ARM-CPU's
https://twitter.com/alexeyab84/status/1436377831974506496
Is not the same scale cut out used in Scaled Yolov4 and previous research? (Though this one is on AP not mAP). It seems it is needed since the improvement would be hard to appreciate if it would show a 0-100 scale, since improvement in OR models are precisely not dramatic.
https://openaccess.thecvf.com/content/CVPR2021/papers/Wang_S...
I honestly find that deceiving and I would’ve said so in a review. Since it /is/ a percentage, I think you should absolutely show the zero - and the maximum if applicable. The correct scale is 0-1. It being a minute difference is exactly the type of thing a fair plot would reveal.
However, I think it’s better to use a semilog plot here - and show 1 - AP. I’m sure there’s a name for that number too. Then perhaps invert the Y axis if you really want the curve trending upward. No need for the cut, and infinite precision.
You of course should note this in the figure caption. This goes doubly so for cutting axes.
There's a reason to not compare it with the YOLOv4 family?
If I recall correctly, the advantage of YOLOv5 over YOLOv4 is still disputed and YOLOv4-tiny seems widely used.
I went to a technical high school in Mexico. Here's divided as 6 years of grade school, 3 of Middle school and 3 of High school.
After finishing this "bachelor" high school you get a technician degree. Mine was an electronics technician. And well, I met some of my best friends there and equals even when most of them went to different majors in college.
I think is great when you can get specialized subjects in such environment since people with similar interests can interact between them more easily.
Granted, there were bullied people, others that just wouldn't fit, but for the most of us (on classrooms over 40 students) we went along fine.
And certainly I felt it a better environment than middle school (12-15yo). Friends that really didn't like it, weren't that interested in electronics or science, but still we got along.
But well, in the end, Mexico is a place where we care about our relationships. Maybe too much.
I don't think the concept of high school is wrong. On any country contacts are important and usually change your life, and any social hub in life is a chance. It's just that, environment and motivation matters.
Graphics also matter. My 6yo Inspiron let me develop with 8GB of RAM, last year I bought a new one with Ryzen5 set the very same link setup and even used the same SSD but it would began to freeze or crash...
The APU was reserving 3GB of RAM for the GPU. The next day I went for 8GB more.
Agreed, the last 4 laptops bought by my family, the ones with dedicated graphics are still going strong, but the ones with integrated intel graphics aged poorly despite having better CPU perf.
I don't think you understood what your parent meant. In their case the GPU was detrimental cause it was reducing the ram available to the rest of the system to 5gb.
Windows images that are over 4.3GB(I don't remember the exact max size) most image burning tools will fallback to NTFS on the USB formatting and some motherboards/laptops don't like this. (Even on my rather modern Dell laptop)
The best solution so far I've found is to manually partition the USB into a 1GB FAT32 partition and the rest NTFS. Then you extract the files where needed [0]
I learned this method just last year but it seems it's pretty universal. At least for Win10 ISOs. And I don't have to boot Windows and wait for its updates!
Edit: I just remembered that you might not be on a Windows device, in which case you may well want to download the ISO and use Rufus to make a bootable stick from it.
I'm love/hating this idea because now I want to burn a LiveUSB running Linux with a QEMU Windows VM already set up on it. I can run it live and have Windows 10 "PE" or I can install it to a host and have Windows up and running fairly smoothly without going through the OOBE.
IIRC you used to be able to make a bootable Windows stick on Linux simply by extracting the contents of the ISO onto a USB stick and marking it bootable. Is that still possible now that EFI is everywhere?
Normally, for UEFI boot, you could be able just to create a FAT32 partition and copy the ISO content there. In the past, it was perfectly workable way to do so.
Except for the unfortunate install.wim file. The first releases of Windows 10 were fine, they had it under 4GB, but some of the half-year releases have it grown over 4GB, and FAT32 cannot handle that. Thus all the mitigations you see here.
The Microsoft's USB tool does not use install.wim; it contains install.esd instead. It is basically the same thing, but with different compression, so it is still a bit under 4GB. You can recompress install.wim into install.esd, if you have the inclination ...and a windows machine nearby (dism /export-image).
The official Windows media creation tool is a bit of an abomination. Primarily because it is slow and has very quirky requirements that it doesn't advertise and only checks for at the very last second.
You download the tool and want to create a Windows 10 bootable USB, so you run the executable.
Then you find out that it won't let you choose where to save the temporary ISO file that it wants to download. It's hardcoded to the C:\ drive which can easily be a small SSD that is already almost full. You grumble a bit, proceed to delete/move some files off of C:\ so that it could download the ISO and then restart the tool.
This time the download finishes successfully, but after the download you get an error because it turns out it wants to extract what it downloaded - and of course hardcoded to the C:\ drive. You moan a bit, proceed to delete/move even more files off of C:\ and restart the tool.
Third time's the charm right? Well after waiting around forever, it has once again downloaded the ISO and this time also extracted it. Then it informs you that it can't actually create the bootable USB stick because it needs to be run as an administrator. You scratch your head in disbelief, wondering why instead of giving you an error message it doesn't just launch a privilege escalation prompt - or at least have the privilege requirement defined in its manifest.
The fourth time has to work. You manually run it as administrator, it re-downloads & re-extracts the ISO, and you still get the error message that it needs to run as an administrator. What is going on? You google for answers and realize that you've found the only Windows app in existence that is not satisfied with mere privilege escalation but instead demands you to be running directly as the administrator.
You log out of your regular user account, log in as the administrator, re-download & re-extract the ISO - and then finally the tool is willing to create the bootable USB.
...
On the other hand it is possible to create the Windows 10 bootable USB with just the command line [1] and this can be done without having the ISO on your C:\ drive, without having to extract the ISO files to a temporary location before being copied to the USB, or having to directly log in as administrator. A simple escalated privilege command prompt will do just fine.
Thus I'm left wondering what is going on at Microsoft. Did they assign some random intern to create the Windows 10 installation media tool? It is very incompetently made.
Thus I'm left wondering what is going on at Microsoft. Did they assign some random intern to create the Windows 10 installation media tool? It is very incompetently made.
I get the same feeling with a lot of other things in Win10 (which I'm only using because I'm forced to). IMHO it's a sign of the shift to a metrics-driven culture, where the "unquantifiable" parts of quality are being completely ignored in favour of bettering numerical ones. Microsoft isn't the only one, this seems to be happening industry-wide.
If you go to the Windows 10 download page, you can trick it into giving you the ISO by changing your responsive view mode in dev tools to an iPad or something and refreshing the page. Then burn with Rufus.
Rufus is vastly more reliable, flexible, and orders of magnitude faster.
So long as you enable updates (to enable online functionality), the "SELECT" button has a dropdown to change it to "DOWNLOAD".
Click "DOWNLOAD", it walks you through a script ("Fido") that grabs the latest Windows 10 ISO (and lets you download it directly or using a browser). You can then use that ISO to make your bootable USB.
I think I did that, then have problems with the windows installer. A driver prompt. I regularly have to give services to laptops, so I prefer options that "just works". This is also the reason I don't usually use the Windows installation media creator, because is slow and won't burn my several ISOs. There was a time the new Win10 iso installation process would fail on any configuration I would try... I lost somewhat around THREE DAYS trying to solve each f*"** cryptic error message and an old iso did the trick. Just a few hours updating and it was done.
Sadly most of my clients won't just accept the option of installing Linux ;)
That seems far more complicated than just using DISM, PowerShell or wimlib (on Linux or Windows) to split the install.wim file into two parts, which I believe older versions of Rufus (or a similar utility) offered to do.
I used the same method as you, but from different source.
I tried to dig into my documentation for the link that I saved, it doesn't seem like I saved the google search terms for it, but they can be deduced from the link anyway.
