ru-vid.com is an unofficial comma.ai and George Hotz video archive. Previously it was named commaai archive. Here you will find live-streams, live coding from ru-vid.com and twitch.tv/georgehotz. Our goal for this channel is to capture every broadcast and make it accessible to everyone. We express gratitude to George Hotz for being so open about his companies. Without George streaming and enabling replay after streaming there is nothing to archive. We appreciate it! All uploaded videos are licensed by George Hotz and comma.ai. We just archive them. We hope you enjoy watching George's videos as much as we do. We are not affiliated with comma.ai or George Hotz. To contact them use comma.ai or tinygrad.org website. Let us know how we can improve your experience on this channel. You can email us at commaaiarchive@protonmail.com if you have any video or audio files about George or comma that are appropriate for sharing and we will consider uploading them.
As far as I can tell what we have done was (up to @5:35:11): -------------------- "orb features" are used to extract keypoints to track from the image. Descriptors are computed for the extracted orb features so they can be compared. extracted orb features are matched using a KNN matcher. In other words, for the last two frames, the descriptors for the keypoints are compared and using a KNN classifier the most similar two keypoints (extracted orb features) are matched. Matched here means that the keypoints from the last two frames should ideally be corresponding to the same part of the image. The matched keypoints (extracted orb features) are then filtered. To do so the RANSAC algorithm which estimates the Fundamental Matrix (F) by ignoring the outliers in the extracted orb features is used. Then using the estimated Fundamental Matrix we calibrated our camera (intrinsic matrix K encodes the camera calibration information). Once the camera is calibrated we use the RANSAC algorithm to estimate the Essential Matrix (E) by ignoring the outliers in the extracted orb features. Extracted the rotation (R) and the translation (t) matrices from the Essential Matrix using SVD and a few other equations. We triangulate points using the keypoints from the last 2 frames and their corresponding rotation and translation (R t) matrices which we had previously obtained. This gives us the 3D homogenious coordinates of the detected orb features in the world. We also keep track of the camera's relative position in the world between the frames of the video by multiplying the calculated rotation and translation matrices (R t) with the pose of its previous frame. We store the computed camera position and the 3D keypoint positions (extracted orb features) in every frame in a python object named mapp. We use this mapp object and the pangolin python library to draw the 3D map on the screen.
George Hotz may be one of the last people on earth that i trust to challenge any new or controversial scientific topic that needs further scrutiny to nail down a true and believable statement and proclamation regarding its fundamentals. also - x- could Georges keyboard be ANY louder ?!?!!
i dont know much about coding so with this for what i understand avoids bugs of the gpu ??? but like the if u get a problem who do u go to if not the gpu vendor ??
!!!! READ THIS, THIS MIGHT SAVE YOU AS A CODER, DON'T MAKE THE SAME MISTAKE THAT I DID !!!!!!!! At 14:50 finally someone who thinks the things in the same way as i do . what genuienly interests me is technology, science & machines (computers) but the software job/career i chose (thinking i'd do these things or learn about new machines or exciting stuff related to computers) doesn''t actually reflect things that genuienly amaze me. that is why i chose to learn/go deep into mathematical basics of ml/ai rather than learning a thousand different (web dev) frameworks/clouds which practically do the same stuff. That said i doesn't mean these things are not useless, i'd use react if i want to build a frontend but the thing here that pisses me off is "the why? ". if your "why " = i need a job to make my living. and it is a software job. what am i doing ? -> someone needed a product and my company is building it and i'm just a cog in the machine that just builds it. don't get me wrong i doesn't imply that this would not be exciting for you and you wouldn't learn new things (may be it gives you enough dopamine to keep you in the game). but what my ultimate concern is, If you happen to do things in the routine software engg. way your intellect has lost its purpose that is the reason ai can replace you, don't you see/get it?. For example, you can become a faang engineer with high pay but you can't become george hotz (using his name just for the example sake, not exaggerating him or degrading faang eng. _ understand the difference). If you still don't understand, see here: lets say you start learning about computers and networks. if this fascinates you initially and start going deep and deep at one point you may notice things like " ah the firmware in this router sucks, i can easily alter things here and play with all the insides of this network". Yes!! that's hacking. and that feels amazing and rewarding because there is some proof of your learning. And there is the other way (mostly people fall into this trap, that includes me , that is the reason that i'm writing this may be atleast one programmers gets saved), you start learning/digging these things because these things interest you but you deviate into a bigger scheme of rat race of job and end up not dirtying your hands with the things you learnt/excited about. The reason this trap is soooo dangerous is 1. this is so hard to notice 2. when you notice (by looking at people like this george hotz) you (me in this case) knew that both the amazing person that made me/you realise and you/me started off not so different, it is just that this person got his hands dirty and i just moved on in the other way thinking that i found a job that i like to do and wasted my absolute potential. AVOID THE TRAP OF LEARNING AND START GETTING YOUR HANDS DIRTY ASAP (Learning how to code ❌️. Doing code / Coding ✅️) understand the nuance & context based on this discussion. Don't be dumb and ask how to code without learning. in the end the conclusion of my rant here would be as a coder: I/YOU WOULD WANT TO BE, WHAT WE LOVE TO BE, AS MUCH AS WE CAN BE. not about creating impact or not about getting big (fame/rich/whatever). that rat trap that defines the trash of a software engineering doesn't fullfill any of the above three things most importantly it makes you avoid ' as much as we can be' part, which later makes you feel like ' i had the same potential and i never became like this person' . That said it is now or never. and the second best time to do what you have missed/ what could have fullfille or push your potential is now. Let me know this helped u by liking this comment.
I understand nothing of programming, Ive only written a single IMPORT line for a library in Python but Im watching G. Hotz to get inspired and see whats this all about.
do you remember paul phoenix from freenode? you probably don't, but i do. give him credit, and treat him well (please locate him). if you have means of locating him, please do. and please award him with some work
Amazing stuff, really educational. Im totally new to programming and besides some personal tries and failed experiments in Python, I have 0 experience or insight but Im learning so much from this! Its a huge privilege to have free educational material from George Hotz. If he's reading this I want to him thank you, stay blessed and much love from a total noob fan from Bulgaria
He's given up on AMD 7900 XTX, yet he keeps coming back. This AMD GPU is like the toxic girlfriend with all the red flags who is really hot, that you see on / off, that you know is bad for you but you can't stop yourself. That said, if you can change AMD and they open-source their driver components It will change the industry.