video_id
stringlengths 11
11
| text
stringlengths 361
490
| start_second
int64 0
11.3k
| end_second
int64 18
11.3k
| url
stringlengths 48
52
| title
stringlengths 0
100
| thumbnail
stringlengths 0
52
|
|---|---|---|---|---|---|---|
LfUsGv-ESbc | that I called copy pasting to go Suraj so will from now and we'll call it just Suraj Inge what we also need are the class labels because that's in defined in the cocoa dataset right so these are the class labels let's take those and okay so this T here these are torch vision transforms we're gonna need that so from say so if you don't know torch vision it's kind of an addition to PI torch that | 234 | 269 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=234s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | just helps you with with images and has a lot of data sets and these transforms they're really helpful because so let's call this image because you can you know resize but they have much more like random cropping and rotating images and so on pretty much everything you need for pre-training and this here is just the standard image net I believe the image net normalization so these are the | 269 | 294 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=269s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | means and these are the standard deviations from the image net data set and let's already resize our image actually to this weight hundred and I believe I believe if you rescale the 640 to 800 you get 600 here right fairly sure okay and then let's display it just because we can okay what it's it's a bit squished but we don't care and let's put that up here so we only need to execute | 294 | 326 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=294s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | it once nice okay so from now on it should be a breeze so what these transforms do is they resize the image okay we don't need that anymore they make it into a tensor and then they normalize by that so if we run our image through this because our image right now is this is pill image right so our our image is this pill image but if we run it through the transforms then we'll get | 326 | 362 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=326s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | a tensor so that's pretty cool so the model as it is a deep learning model that expects batches so we'll unscrew is that in the first dimension and then we get batches so shape let's see we don't have on skis no of course we don't so this is a one image of three channels of 600 by 800 so this is the Y index coordinates I guess are shifted yes in pi torch cool so we'll call this | 362 | 399 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=362s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | our image tensor now we just need to put it into the model so model we put that in there and since we don't let's actually up here put the model in eval mode I don't know if that's already done but you know you can never be sure enough that the batch norms aren't so I think it probably doesn't have batch norms okay you're not utilizing the GPU we'll do that we'll do that | 399 | 433 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=399s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | Thanks so how do we use the GPU we put our model on the GPU model equals model CUDA yes yes yes I think so this is gonna work okay we're gonna come back to this later so we forward our image of course we also need that on the GPU and it's worked did this work this worked nice okay and since this is just for evaluation we should probably go with no grad right here because we don't need | 433 | 480 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=433s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | this whole gradient stuff if we do that okay I'm dumb there you go and nothing happens of course because we need to capture the output somehow let's look at that output Wow Wow just wow so the output is a dictionary right because we get back class labels and bounding boxes so let's look at the bread boxes let's look at that tensor that's a tensor very nice let's look at its shape let's not print | 480 | 521 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=480s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | giant tensors anymore cool so since this was a batch of one we should probably go the zeroeth and you can see right here there is a hundred bounding boxes and each one has four numbers and if you go with the other thing that's in there the log it's then you'll see that there are also should be a hundred log it's and hello there should be a hundred log it's and each one is of size 92 because there | 521 | 556 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=521s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | are 92 different classes 92 we'll see about that well one is going to be the nothing class right by the way how many classes do we have we have 91 classes okay cool we can deal with that all right so what are we gonna do next what we want to do is for each of the for each of the for each of the log it predictions we want to find which classic corresponds to so | 556 | 596 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=556s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | what we're going to do is we're going to take the Arg max of the last dimension right so you can see here almost all of these things correspond to class 91 and class 91 is not in our classes because our class is only length 91 so that must be the nothing class so what we can technically do is for log its and boxes in let's just zip them together and [Music] like this okay class is oops | 596 | 639 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=596s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | class as the law gets Arg max if that's 92 or let's say safe that's larger than the length of our classes we'll just skip it for now okay so that should work somehow and if not then our label should be the class index right here so let's just see what the detector detects right here it detects nothing why does it detect nothing that's isn't seem good what are we doing wrong we zip together | 639 | 701 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=639s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | the log it's oh yeah of course we still need the zero with entry we are dumb dumb dumb cool so so so so we can delete this and now finally beautiful dogs - dogs detected excellent so now for each of these dogs we want the bounding box okay so now we somehow need to think of how are we gonna draw this on an image and well let's let's actually make a copy of that image because I don't really trust | 701 | 750 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=701s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | myself and then at the end of this we're just going to display that image right now actually the reason I make a copy is because in these in this pillow library you can actually draw on these images and we're going to that to draw these bounding boxes so for that we need an image draw if I remember correctly and I think later we also want some text so we need an image font yes | 750 | 781 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=750s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | all right so let's draw a bounding box right here where so first of all let's look at that bounding box let's call this box box print box dot shape and break right here what's happening let's not do this right now so this is a boxes of size four now this could be two things it could be X 0 y 0 X 1 Y 1 so the two corner points or the kind of the boundaries or it could be X Y width | 781 | 827 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=781s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | height now from the paper I know that they predict the center and the width and the height so I'm gonna go with that and I'm just gonna guess that it's like X Y WH and not some other way around if this is a bad guess then yeah we'll see we can just print out one of these boxes and honestly that looks reason oh by the way we should scale that up yeah so these are normalized coordinates | 827 | 855 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=827s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | probably between 0 and 1 so we should scale that up so we should probably the x coordinates which is scaled by 800 and the Y by 600 so let's do it so first of all we scale our box by 800 in the X and here is a Y and the width is the X direction and this is the Y Direction boom okay we should probably get that on CPU will just hack together a bunch of things right here ok so now this isn't | 855 | 893 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=855s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | the correct so we sold our x and y and WH are going to be this box so now we need to actually draw on the image we're gonna do that so let's first go X 0 X 1 is X minus W 1/2 X plus W half y 0 y 1 is the same for a y with H plus H half Coolio now we need an image draw object so I think draw on this image so whatever you draw on the draw object will end up on | 893 | 939 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=893s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | the image so we can use that to draw a bounding box and let's just quickly look it up so pill Python draw rectangle maybe there we go okay so there's this rectangle yeah there's the rectangle function and you can see you put in a shape XY here and width height like this wait for real we wouldn't even have to need to transform it I'm pretty sure you can go X I thought I remember you could | 939 | 978 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=939s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | do the different thing as well but it's called rectangle okay so let's do that so draw rectangle and we'll go we'll go X 0 or we'll go X Y width height let's display that down here yeah that looks that looks nothing like we want but it's you know it's a start maybe actually we need the other thing here we need X 0 y 0 X 1 Y 1 mm yes yes doggy okay we still have the break in here | 978 | 1,032 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=978s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | now we get both dogs nice nice okay let's do I think Phil yes red and let's go for with five or so five seems like a good width oh god five is a terrible with oh it's not feel I think it's its outline yeah yeah okay okay let's go still go with five cool we got our dogs now we need to put like some some snappy text labels I think there's actually a pill image draw | 1,032 | 1,077 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1032s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | text I think that exists because I've this font thing yeah exactly so you need the font thing get it font in there and then yeah exactly you could put a text like this okay so you probably need the x and y coordinates of the text so let's do that W dot text and let's just go with x and y right here put it right in the middle and the text is going to be our label of | 1,077 | 1,113 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1077s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | course and we want the fill that's now going to be the color of the text let's go with white and the font we're going to load some font right here font dot how we're doing this true type true type ah no not cheating let's just go with regular fonts it won't look as fancy but we'll be fine so we're where is our text you see it I don't see it red let's make it red yes there we go okay | 1,113 | 1,178 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1113s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | so it wasn't red enough this should work on it so I did we just I just not see it I'm domina cool so we have two dogs how easy was that actually we wasted the most time with like bounding boxes and stuff absolutely cool right okay so now we can have some fun with it I'm going to scale this down for a bit because you don't need to see the actual code anymore so | 1,178 | 1,204 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1178s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | much so you can see the image more so we'll go to the images and the first thing I want to do is the dress what does this think of the dress okay so we'll copy that and we'll go into our collab and just paste this right here butter boom but a beam sounds nice and what is wrong the size of a tensor must match the size of a tensor we do something wrong transform image or images this | 1,204 | 1,260 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1204s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | maybe this is like an RGBA image I think if this is rgba we should just convert it to like an RGB pretty sure you can do something like this right here this should work as an alpha Channel then that will remove it yes now it works okay let's see what the model thinks of this yeah okay apparently there's a car and there's a surfboard and there's a person and there's a person nice see well we didn't | 1,260 | 1,304 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1260s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | figure out whether the dress was blue or white through gold it was just a person now they you could actually like threshold by how sure you are of a given class but where's the fun in that so let's go further and let's do some Rorschach inkblots because those are always lots and lots of fun so which one should we go for it this one looks like fun okay so we'll put this into here and | 1,304 | 1,351 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1304s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | it's astonishing right it's this cocoa data said it only has these 90 classes like it doesn't have anything anything else so it's a cake it's a cake and this here what is it okay we'll have to go maybe with blue what is it stop sign okay but so you might think it what if what if we want more like what if we want more predictions so there is a hack right right now the model can always | 1,351 | 1,383 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1351s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | assign math to this not a class thing like right here this class 91 in order for it to say I don't think there's anything there but generally we have a hundred predictions right so you see where this is going so yes let's let's change it but let's change it up a bit and let's go here let's first extract these tensors and boxes okay so we have the boxes and this | 1,383 | 1,424 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1383s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | and log its and boxes okay so we got that what we wanna do is basically we want to filter the we want to basically just remove the last class before we do the Arg max and thereby we want to force the model to make a prediction now it won't be a very good prediction because of course this is only the second highest class and it's arguable how much that counts but still it will do | 1,424 | 1,457 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1424s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | something so this must be done in the log it's right so well look at the log it's and the log it's our of shape 100 so we have 100 predictions of 92 classes now the first thing we want to do is just remove the last class so let's go everything here until the last class all right so now we have 91 actually let's make it more generic whatever thing however many classes are okay so we | 1,457 | 1,489 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1457s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | don't have this class anymore so now if we do the softmax over the last thing we can technically we get 91 but now they're normalized so they add up to one so it's kind of a probability distribution next we we want to find the max over this and that that will give us a max output so we don't want to plot all the 100 predictions because that would just be like like squares all over | 1,489 | 1,524 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1489s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | the place and we'd have no clue what happening so this max output right here this what we're trying to find is we're trying to find a let's say the five best predictions or so the five ones where the model thinks where the model is most confident it's not really good metric but you know so these are the probability values of all of the hundred predictions so what we want is like the | 1,524 | 1,558 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1524s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | top K okay so let's go with five and again we'll get like a top K output let's call that top K and I think it also has like values and indices yes so now we simply need to filter from the log it's and the boxes where these these top ones are so well filter the log it's [Music] will filter the log it's by that top K indices and we'll also filter thee I am not very gifted today boxes | 1,558 | 1,622 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1558s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | by the way I'm using a collab just because it's nice to kind of play around with a model because if I were to use a file I'd have to restart reload the model over and over again just not as nice so now we have the log it's and the boxes and if we do that right now we get always the top 5 predictions how nice is that and you can see the top 5 predictions are probably still kkkkkk | 1,622 | 1,649 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1622s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | cake and just to verify that and we can put its shape yeah see this is what I don't like about this stuff yes okay so we just have five predictions of 92 things and we don't want the 92 we've already said so we just want the 91 let's actually could put that here [Music] okay so now we have five by 91 and now to give us the top five are there we go so many takes and many stop sighs that's | 1,649 | 1,707 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1649s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | fine that's cool so the ultimate test right here is going to be yes the human adversarial example let's check it out so we put in a Jackson Pollock image and we'll see what the model says now we're actually forcing it to make predictions right so it can't escape it will need to do something okay I made another mistake I would need to copy the image address right here like this | 1,707 | 1,751 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1707s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | that's what happens when you're not an idiot you get the actual image so what does the model think of our pretty image okay can't even read that so let's make this into white bird bird bird okay lots of birds in this image clearly clearly lots of birds in this image let's try another one let's go with this this this one yes yes absolutely love it love it okay so we copy image | 1,751 | 1,796 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1751s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | address and beam Mormons Wow there's a lot of birds in these Pollock images just so many birds okay let's try one last how about this one this one is a bit more human-friendly right put it in here and and and okay we get some detections there's a clock right here there is a what's that how's horses let's print let's print the labels so just so we know what they are | 1,796 | 1,869 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1796s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | cake horse car horse and clock okay so I see the clock like this here is clearly a clock then this rectangle on the right side must be something let's put this to read as well now that's terrible ah white back to white how about back to white okay clock we got horse right here and house probably and the entire image is again a cake yes okay so as you can see it is a pretty pretty | 1,869 | 1,923 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1869s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | good system but of course it is only these 90 classes but it's for now it's a it's pretty cool and it works pretty well and just the easiness with which you get which which you can get this stuff elephants in Kruger National Park just the easiness is astonishing you can just load it up kind of have this have a bit of a notebook and with a bit of like a very few lines of code you can put something | 1,923 | 1,959 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1923s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
LfUsGv-ESbc | together that detects these bounding boxes lots of elephants and remember we only have the top five elephants right here so what happens if we go for more where is our top k so here we can let maybe say the top 15 predictions and as always if we want to make the model to make its own decision we can simply revert back and add back the no class label all right with that I hope you | 1,959 | 1,988 | https://www.youtube.com/watch?v=LfUsGv-ESbc&t=1959s | [Code] How to use Facebook's DETR object detection algorithm in Python (Full Tutorial) | |
r7cYsgB4G1s | Let's imagine for a few moments what our life would be like if we could access let's say 20% of our brain's capacity If you want to have something show up in your life The kind of person you would like to become manifest something new into your life something powerful, whatever it might be You obviously must first be able to imagine it Your imagination is able to do all that you ask in | 0 | 35 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=0s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | proportion to the degree of your attention So what kind of attention do you place on your desires? Einstein's most famous quote one of his most famous observations. He said imagination is more important than knowledge Knowledge is limited Imagination encircles the world Logic will get you from A to B, but imagination will take you everywhere Make your future dream a | 35 | 79 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=35s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | present fact by assuming the feeling of the wish fulfilled That which you feel yourself to be you are and You are given that which you are. So assume the feeling that would be yours were you already in possession of your wish and your wish must be realized so live in the feeling of being the to be and that you shall be if this assumption about what you would like to become is | 79 | 118 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=79s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | persisted in until it becomes your dominant feeling the attainment of your ideal is Absolutely inevitable You must first assume the feeling of a wish fulfilled in all aspects of your life Don't allow anybody elses opinions Don't allow what it says on the internet. Don't allow the research. Don't allow what anybody out there tells you is possible or not possible | 118 | 149 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=118s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | for you if you advance confidently in the direction of your own dreams and Endeavor to live the life, which you have imagined You will meet with a success unexpected in common hours it will chase after you if you can place into your Imagination what it is that you would like to attract and begin to feel it Start retraining your subconscious mind and your subconscious mind | 149 | 186 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=149s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | it responds to what it is that you suggest to it The subconscious mind moves your life 96 to 97 percent of everything that you do is done as a result of your subconscious mind And when your subconscious mind gets programmed it goes ahead and Respond to whatever it is. Your conscious mind has placed into it You are the creator this is the mystery This is the great secret known by the seers and prophets and mystics throughout the ages. This is the truth that you can never know | 186 | 236 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=186s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | intellectually Many of you as I have been as I am are where you are in your life Based upon what you believe and it's not just what you think you believe on the surface It's also your shadow beliefs that are holding you back from moving into the life that you believe You deserve What I know is if you're not looking at the shadows if you're not looking at what is | 236 | 268 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=236s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | Subconsciously running through the tape in your mind telling yourself. You're not good enough. You're not worthy enough. You're not smart enough You're not enough which is a tape that's playing for a lot of people If you're not conscious of that then you end up acting out of that Belief system and not out of what you know to be the truest or want to be the choice for yourself | 268 | 294 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=268s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | You are where you are today in part because of what you've been saying about yourself Words are like seeds when you speak something out. You give life to what you're saying if you continue to say it Eventually that can become a reality you Are planting seeds when you talk at some point you're going to eat that fruit. My challenge is make sure You're planning the right kind of seeds if you want apples | 294 | 325 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=294s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | You have to sow apple seeds if you want oranges You can't plant cactus seeds poison ivy seeds mushroom seeds You're going to reap fruit from the exact seeds that you've been sowing in other words You can't talk negative and expect to live a positive life You can't talk defeat and expect to have victory you can't talk lack You're not enough can't afford it Never get ahead and expect to have abundance if you have a poor mouth. You're going to have a poor life | 325 | 362 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=325s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | And this is great when we're saying things like I'm blessed I'm strong I will accomplish my dreams I'm coming out of there That's not just being positive. You are prophesying victory Prophesy and success prophesy new levels and your life will move in the direction of your words But too many people go around prophesy on just the opposite. I never get any good brights. I'll never get back in shape | 362 | 391 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=362s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | Business is slow. I'll probably get laid off Flu season is here. I always get it. They don't realize they are prophesy and defeat It's just like they're calling in bad breaks mediocrity lack You don't become what you want because so much of wanting is about Living in the space of what you don't have that's why Jim Carrey's story is so powerful Because he started to act as though he already had it. He would go up to Mulholland Drive | 391 | 422 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=391s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | He would drive away sing thinking. I already have those things I just haven't accessed them as yet I believe Those things are going to come to me and I'm going to act like they are so I'm gonna move forward in my life in order to Draw that to myself in such a way that my actions are in alignment with what I say, I believe So if you start to think about that really why are you where you are in your life the choices that you have made? | 422 | 450 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=422s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | Have been because of what you believe to be true for yourself The time is now the time is now to express and for people to believe in themselves The time is now for it to be okay to be great People in this world shun people for being great for being a bright color for standing out but the time is now to be okay to be the greatest you You can talk yourself out of your destiny | 450 | 485 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=450s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
r7cYsgB4G1s | Negative words can keep you from becoming who you were created to be don't fall into that trap Quit calling in defeat quit talking about how it's not going to happen You should wipe down your Write down what you want to see happen in life Any areas that you're struggling in where you need to improve write it down like it's already done and then every day declare that decree | 485 | 513 | https://www.youtube.com/watch?v=r7cYsgB4G1s&t=485s | YOU ARE THE CREATOR | Warning: This might shake up your belief system! Morgan Freeman and Wayne Dyer | |
T7o3xvJLuHk | convolution is a measure of overlap between two functions as one slides over the other mathematically it's a sum of products the standard convolution operation is slow to perform however we can speed this up with an alternative method that is the topic of this video depth wise separable convolution let's first very quickly go over the basics of convolution on an input volume consider | 0 | 28 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=0s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | an input volume f.o shape d f cross d f cross m where DF is the width and height of the input volume and M is the number of input channels if a color image was an input then M would be equal to 3/4 the RG and B channels we apply convolution on a kernel K of shape DK cross DK cross M this will give us an output of shape D G cross DG cross 1 if we apply n such kernels on the input | 28 | 59 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=28s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | then we get an output volume G of shape DG cross DG cross n the convolution operation takes the sum of products of the input and the kernel to return a scalar this operation is continued by sliding the kernel over the input I've explained this concept in detail on my video on convolution neural networks check that out for a clear understanding I'm more concerned now with the cost of | 59 | 88 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=59s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | this convolution operation so let's take a look at that we can measure the computation required for convolution by taking a look at the number of multiplications required so why is that it's because multiplication is an expensive operation relative to addition so let's determine the number of multiplications for one convolution operation the number of multiplications | 88 | 116 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=88s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | is the number of elements in that kernel so that would be D K times D K times M multiplications but we slide this kernel over the input we perform DG convolutions along the width and DG convolutions along the height and hence D G cross DG convolutions over all so the number of multiplications in the convolution of one kernel over the entire input f is DG square times D K | 116 | 146 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=116s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | square times M now this is for just one kernel but if we have n such kernels which makes the absolute total number of multiplications become n times D G square times D K square times M multiplications let's now take a look at depth wise separable convolutions in standard convolution the application of filters across all input channels and the combination of these values are done | 146 | 175 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=146s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | in a single step def y separable convolution on the other hand breaks us down into two parts the first is depth wise convolution that is it performs the filtering stage and then point wise convolution which performs the combining stage let's get into some details here depth wise convolution applies convolution to a single input channel at a time this is different from the | 175 | 201 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=175s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | standard convolution that applies convolution to all channels let us take the same input volume F to understand this process F has a shape D F cross D F cross M where D F is the width and height of the input volume and M is the number of input channels like I mentioned before for depth wise convolution we use filters or kernels K of shape DK cross DK cross one here | 201 | 230 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=201s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | DK is the width and height of the square kernel and it has a depth of 1 because this convolution is only applied to a channel unlike standard convolution which is applied throughout the entire day and since we apply one kernel to a single input channel we require M such DK cross DK cross one kernels over the entire input volume F for each of these M convolutions we end up with an output | 230 | 258 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=230s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | DG cross DG cross one in shape now stacking these outputs together we have an output volume of G which is of shape DG cross DG cross M this is the end of the first phase that is the end of depth wise convolution now this is succeeded by point wise convolution point wise convolution involves performing the linear combination of each of these layers here the input is the volume of | 258 | 292 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=258s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | shape DG cross DG cross M the filter K PC has a shape one cross one cross M this is basically a 1 Cross 1 convolution operation over all M layers the output will thus have the same input width and height as the input D G cross DG for each filter assuming that we want to use some n such filters the output volume becomes D G cross DG cross n so that's great we got this down now let's | 292 | 329 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=292s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | take a look at the complexity of this convolution we can split this into two parts as we have two phases first we compute the number of multiplications in depth wise convolution so here the kernels have a shape DK cross D K cross 1 so the number of multiplications on one convolution operation is all DK times DK DK square when applied over the entire input channel this convolution is performed DG | 329 | 358 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=329s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | x DG number of times so the number of multiplications for the kernel over the input channel becomes DG square times DK square now such multiplications are applied over all em input channels for each channel we have a different kernel and hence the total number of multiplications in the first phase that is depth wise convolution is M times D G square times D K square next we compute | 358 | 391 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=358s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | the number of multiplications in the second phase that is point wise convolution here the kernels have a shape one cross one cross M where m is the depth of the input volume and hence the number of multiplications for one instance of convolution is M this is applied to the entire output of the first phase which has a width and height of D G so the total number of | 391 | 416 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=391s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | multiplications for this kernel is d G times D G times M so for some n kernels will have n times D G times D G times M such multiplications and thus the total number of multiplications is the sum of multiplications in the depth wise convolution stage plus the number of multiplications in the point-wise convolution stage we can take M times D G squared common now we compare the | 416 | 447 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=416s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | standard convolution with depth wise convolution we get the ratio as the sum of reciprocal of the depth of output volume that is n and the reciprocal of the squared dimensions of the kernel DK to put this into perspective of how effective depth wise convolution is let us take an example so consider the output feature volume n of 1024 and a kernel of size 3 that's DK is equal to 3 | 447 | 477 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=447s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | plugging these values into the relation we get zero point 1 1 2 in other words standard convolution has 9 times more the number of multiplications as that of depth Y separable convolution this is a lot of computing power we can also quickly compare the number of parameters in both convolutions in standard convolution each kernel has k times D K times M learn about | 477 | 504 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=477s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | parameters since there are n such kernels there are n times M times D K squared parameters in depth by separable convolutions will split this once again into two parts in the depth wise convolution phase we use M kernels of shape DK cross DK in point wise convolution we use n kernels of shape 1 Cross 1 cross M so the total is M times DK square plus M times n or we can just | 504 | 536 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=504s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | take M common taking the ratio we get the same ratio as we did for computational power required so we understood exactly what depth wise convolution is and also its computation power with respect to the traditional standard convolution but where exactly has this been used well there are some very interesting papers here the first is on multi model neural networks these | 536 | 566 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=536s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | are networks designed to solve multiple problems using a single network a multi model network has four parts the first is modality Nets to convert different input types to a universal internal representation then we have an encoder to process inputs we have a mixer to encode inputs with previous outputs and we have a decoder to generate outputs a fundamental component of each of these | 566 | 593 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=566s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | parts is depth wise separable convolution it works effectively in such large networks next up we have exception a convolution neural network architecture based entirely on depth wise separable convolution layers it has shown the state-of-the-art performance on large datasets like Google's jft image data set it's a repository of 350 million images with 17,000 class labels | 593 | 622 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=593s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | to put this into perspective the popular image net took 3 days to Train however to Train even a subset of this jft data set it took a month and it didn't even converge in fact it would have approximately taken about three months to converge how'd they let it run to its full length so that's useful this paper is pushing convolution neural networks to use depth Y separable | 622 | 647 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=622s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | convolution as the de facto up third we have mobile Nets a neural network architecture that strives to minimize latency of smaller scale networks so that computer vision applications run well on mobile devices mobile nets used F Y separable convolutions in its 28 layer architecture this paper compares the performance of mobile nets with fully connected layers versus depth wise | 647 | 673 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=647s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | separable convolution layers it turns out the accuracy on image net only drops a 1% while using significantly less number of parameters from twenty nine point three million the number of parameters it's down to just 4.2 million we can see the mulch as the number of multiplications and additions which is a direct measure of computation has also significantly decreased for depth by | 673 | 700 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=673s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | separable convolution mobile Nets so here are some things to remember in this video depth Y separable convolution decreases the computation and number of parameters when compared to standard convolution second is that depth Y separable convolution is a combination of depth wise convolution followed by a point wise convolution depth wise convolution is the filtering step and | 700 | 725 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=700s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
T7o3xvJLuHk | point wise convolution can be thought of as the combination step finally they have been successfully implemented in neural network architectures like multi model networks exception and mobile nets and that's all I have for you now thank you all for stopping by today if you liked the video hit that like button if you want to stick around hit that subscribe button | 725 | 747 | https://www.youtube.com/watch?v=T7o3xvJLuHk&t=725s | Depthwise Separable Convolution - A FASTER CONVOLUTION! | |
Z6rxFNMGdn0 | the following is a conversation with Ian good fellow he's the author of the popular textbook on deep learning simply titled deep learning he coined the term of generative adversarial networks otherwise known as Ganz and with his 2014 paper is responsible for launching the incredible growth of research and innovation in this subfield of deep learning he got his BS and MS at | 0 | 26 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=0s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | Stanford his PhD at University of Montreal with yoshua bengio and Erin Kerrville he held several research positions including an open AI Google brain and now at Apple as the director of machine learning this recording happened while Ian was still a Google brain but we don't talk about anything specific to Google or any other organization this conversation is part | 26 | 52 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=26s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | of the artificial intelligence podcast if you enjoy it subscribe on YouTube iTunes or simply connect with me on Twitter at lex friedman spelled fri d and now here's my conversation with Ian good fellow you open your popular deep learning book with a Russian doll type diagram that shows deep learning is a subset of representation learning which in turn is a subset of machine learning | 52 | 79 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=52s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | and finally a subset of AI so this kind of implies that there may be limits to deep learning in the context of AI so what do you think is the current limits of deep learning and are those limits something that we can overcome with time yeah I think one of the biggest limitations of deep learning is that right now it requires really a lot of data especially labeled data there's | 79 | 104 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=79s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | some unsupervised and semi-supervised learning algorithms that can reduce the amount of labeled data you need but they still require a lot of unlabeled data reinforcement learning algorithms they don't need labels but they need really a lot of experiences as human beings we don't learn to play pong by failing at pong two million times so just getting the generalization ability better is one | 104 | 126 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=104s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | of the most important bottlenecks and the capability of the technology today and then I guess I'd also say deep learning is like a of a bigger system so far nobody is really proposing to have only what you'd call deep learning as the entire ingredient of intelligence you use deep learning as sub modules of other systems like alphago has a deep learning model | 126 | 152 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=126s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | that estimates the value function most reinforcement learning algorithms have a deep learning module that estimates which action to take next but you might have other components here basically as building a function estimator do you think it's possible you said nobody is kind of in thinking about this so far but do you think neural networks could be made to reason in the way symbolic | 152 | 175 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=152s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | systems did in the 80s and 90s to do more create more like programs as opposed to functions yeah I think we already see that a little bit I already kind of think of neural nets as a kind of program I think of deep learning as basically learning programs that have more than one step so if you draw a flowchart or or if you draw a tensor flow graph describing your machine | 175 | 201 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=175s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | learning model I think of the depth of that graph is describing the number of steps that run in sequence and then the width of that graph is the number of steps that run in parallel now it's been long enough that we've had deep learning working that it's a little bit silly to even discuss shallow learning anymore but back when I first got involved in AI when we used machine learning we were | 201 | 220 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=201s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | usually learning things like support vector machines you could have a lot of input features to the model and you could multiply each feature by a different weight but all those multiplications were done in parallel to each other there wasn't a lot done in series I think what we got with deep learning was really the ability to have steps of a program that run in sequence | 220 | 239 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=220s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | and I think that we've actually started to see that what's important with deep learning is more the fact that we have a multi-step program rather than the fact that we've learned a representation if you look at things like res nuts for example they take one particular kind of representation and they update it several times back when deep learning first really took off in the academic | 239 | 264 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=239s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | world in 2006 when Geoff Hinton showed that you could train deep belief networks everybody who was under ested in the idea thought of it as each layer learns a different level of abstraction but the first layer trained on images learn something like edges and the second layer learns corners and eventually you get these kind of grandmother's cell units that recognize | 264 | 283 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=264s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | specific objects today I think most people think of it more as a computer program where as you add more layers you can do more updates before you output your final number but I don't think anybody believes the layer 150 of the resin it is a grand grandmother cell and you know layer 100 is contours or something like that okay so you think you're not thinking of it as a singular | 283 | 308 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=283s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | representation that keeps building you think of it as a program sort of almost like a state the representation is a state of understanding and yeah I think of it as a program that makes several updates and arrives it better and better understandings but it's not replacing the representation at each step its refining it and in some sense that's a little bit like reasoning it's not | 308 | 332 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=308s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | reasoning in the form of deduction but it's reasoning in the form of taking a thought and refining it and refining it carefully until it's good enough to use do you think and I hope you don't mind we'll jump philosophical every once in a while do you think of you know a cognition human cognition or even consciousness as simply a result of this kind of cincuenta sequential | 332 | 357 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=332s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | representation learning do you think that can emerge cognition yes I think so consciousness it's really hard to even define what we mean by that I guess there's consciousness is often defined as things like having self-awareness and that's relatively easy to turn into something actionable for a computer scientists the reason about people also defined consciousness in terms of having | 357 | 380 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=357s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 | |
Z6rxFNMGdn0 | qualitative states of experience like qualia and there's all these philosophical problems like could you imagine jambe who does all the same information processing as a human but doesn't really have the qualitative experiences that we have that sort of thing I have no idea how to formalize or turn it into a scientific question I don't know how you could run in | 380 | 401 | https://www.youtube.com/watch?v=Z6rxFNMGdn0&t=380s | Ian Goodfellow: Generative Adversarial Networks (GANs) | Lex Fridman Podcast #19 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.