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
|
|---|---|---|---|---|---|---|
Uumd2zOOz60 | here and I so for I would input this sequence right here and then to detect an object I would sort of think that maybe the Bert you know Bert has an output that is the same length as the input right so it's it's very good at sequence tagging and things like this so maybe how it detects an object is going to be that it sort of like tags the tags the center location in the pixel of an | 779 | 809 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=779s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | object right here or a tag somehow the corners of the of the bounding box but then I don't know how this is going to be in parallel maybe Bert outputs like a score for each location and then you do some kind of matching right here so this is my initial hypothesis of what's going on and then I scroll through and honestly the first thing I do is I go and find the pictures and know no | 809 | 835 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=809s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | different in all like since since you first book you read that's what you do I go and find the pictures because usually if someone proposes anything new that they're gonna try to make a picture of it luckily I don't do like super theoretical what not your Bayesian generalization bounds and I don't know so most often papers I read have some sort of picture and that's very helpful | 835 | 862 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=835s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | to me I know I know but yeah so I find this picture and here I see okay you have image you have CNN okay gives you a set of image features or so far so good then transform or encoder decoder then set of box predictions so all of them come out here and I already read they're in parallel and then bipartite matching laws so here they I can see they color these in different ways and these color | 862 | 891 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=862s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | appear to match with these colors right here right in the green here and these they they also this is a very good graphic but from this I can already read that these here go to the no object a lot of times the graphics aren't very good so this this is where I'm not saying in every paper you can learn by looking at the graphics like sometimes the graphics are terrible and you're | 891 | 915 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=891s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | like what's going on here I like I don't this this makes no sense this happens a lot in this paper right here this happens to be very very good explanatory graphics so I'll take advantage of that and I do the same thing in the other papers right but then later when it doesn't match what I read in the text I'll have to you know update my belief and so on but here I see that these go | 915 | 940 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=915s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | to no object and this goes to no object so I don't know yet that this is the test set at the point where I read this I was sort of confused by this but I recognized that each of these boxes right here is going to be either resulting in a bounding box or in the no object prediction so from that I could conclude that these things here are maybe some sort of a fixed set right but | 940 | 971 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=940s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | I still thought that you know these that this would actually be the output of these image features so that in this case you'd have like six set of image features and then you'd have like Bert here even though that's not an encoder decoder I still this was still my running hypothesis that somehow you'd map these image features to these boxes right here so and I didn't know what to | 971 | 997 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=971s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | what to make of this this thing right here so then I went through some more and look for more pictures and there are not sometimes I also kind of glanced at the formulas but okay when I Everest I see this this is just I mean this is kind of useless like okay cool you minimize the loss thanks this okay didn't really pay attention to that ah new picture cool so this picture is much | 997 | 1,024 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=997s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | more informative than the other picture yeah I believe with the other picture they were trying to show case this loss how they do the matching and even though I could read a lot from that picture I did not get that part and that therefore I felt when I saw this and I just glanced at it I'm like wait what what's different then up here it seems like the same but okay let's look at | 1,024 | 1,049 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1024s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | this so again we see okay you have set of image features that comes out of the CNN so that conforms with my belief but then this here goes into a transformer encoder and this comes out so immediately I see oh this is not the same as these boxes here right that was my hypothesis that these things here would be the colored boxes so I I say okay obviously that's not what happens this | 1,049 | 1,081 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1049s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | thing here seems to be sort of the encoded image information then that's somehow fed into here and that then there are these object query things and they seem to correspond to this so I'm a bit more confused right now what I can see is that these then will result in these in these boxes okay so being confused by that I look for more pictures so I go look for more pictures | 1,081 | 1,115 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1081s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | and this here seems to be like of a visualization a lot of these papers have some sort of ablation experiments or so and so on this I just find a really cool picture for now I don't know yet what it means this I don't know yet what it means and I go down skip off this and then back here in the appendix I find this here which I immediately map to the previous where this is the anchor and | 1,115 | 1,142 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1115s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | this is a decoder and I've already read the attention is all you need paper and at that point it clicked and means like this is not a Burt transformer this is one of these transformers that has an encoder and the decoder even though they told me like 50 billion times already I was too stupid until this point so now I know okay okay I see what's going on so the image goes through here and then | 1,142 | 1,164 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1142s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | this goes as a side input like as an attention from the decoder to the encoder like I know in NLP right so in NLP this here would be a source sequence like maybe if you do translation and this here would be a target sequence so now whenever I see a transformer like this and it outputs something this I I look at it as okay this here is sort of the input that goes as like a side input | 1,164 | 1,194 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1164s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | over here and usually here you have the target sequence but that's not the case right here right you have these object queries so this is how far I get from the pictures now I go up so I have a sort of I have questions now I have questions and that's when I start reading the paper only now do I start reading the paper after I've looked through all the images form the | 1,194 | 1,221 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1194s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | hypothesis and sort of have questions on how this works and we'll go a bit faster from now on - just not bore you with all the things so the introduction is often very important even though it's called introduction and maybe you know if you read a book like if there's like introduction or a prologue or something like this it's often kinda pointless introduction in these research papers is | 1,221 | 1,247 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1221s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | one of the most important points because all of these papers they try basically all of them try to convince a reviewer to accept them and in order to do that they will set up their main points and their main story immediately in the introduction so what you'll usually have is a problem statement which is here like why what's what's wrong right now and then you have like a story of how | 1,247 | 1,275 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1247s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | their paper addresses the issue okay and that's that's here we streamline the training pipeline by viewing object prediction yada yada yada this is often formulates in words what the paper is about and what contribution the paper makes right this is like a this is like a longer abstract the abstract is often very very cryptic very dense this here is often much more | 1,275 | 1,304 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1275s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | informative of what the paper does so for understanding the paper and a high level the introduction is the best place but given that I've already looked at the images and so on I don't actually draw many new much new information from this thing then is related work and honestly I I skip it like unless I'm the actual reviewer of a paper like when I'm the reviewer of a paper I read the | 1,304 | 1,333 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1304s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | related work but often related work is just like you first of all you site a bunch of your friends and then you cite the mandatory papers and then you cite every single person that you think could be a reviewer because or you've actually been rejected from a conference with a reviewer claiming that your you haven't compared or you haven't cited data or that paper you can pretty much be sure | 1,333 | 1,355 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1333s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | that that's the if if it's not a glaring of may omission if it's like a niche paper and you haven't cited it then you're like okay I'm gonna cite it just because the next conference you could be Mary viewer again so I'm not I'm not sure that these related work sections they're necessary like if someone wants to write their theses and they go and read this paper and they want references | 1,355 | 1,381 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1355s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | oftentimes this is a good place but a lot of it is just blah blah blah blah blah okay I know disagree with me if you want oh yeah - maybe - reading quality so I tend to at this point I tend to not skim so at first I skim but at this point I tend to read every sentence and read it closely and understand it and when I realize like I'm tired or something I don't just skim the paper I've tried to | 1,381 | 1,413 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1381s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | skim papers and it doesn't doesn't work try to read every sentence understand every sentence and okay if you don't understand it don't stop reading because of that but try to not skim and be like oh yeah yeah yeah okay I gotta go to go to get away that is not helpful except related work skip completely cool then a lot of times in this paper now is the the model and this is the section I'm | 1,413 | 1,442 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1413s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | actually interested in right so I read very very closely here and then I find out what their their loss is all about and again I stress read these things and understand them right sometimes it's hard but if you're if you're confused that means you either if they've done a bad job or they made a mistake or that you haven't under stood something if you can't understand | 1,442 | 1,472 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1442s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | the sentence try to read on maybe it's clarified later and then you know go back but again do not do not like just start a lot of times when I read paper previously like I wouldn't understand something quite well yet and then I would be like oh yeah yeah yeah and then I noticed that I start skipping and skimming more and more because that would you know pop up again and again | 1,472 | 1,498 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1472s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | and I wouldn't understand it again and again and then at the end I would just be kind of glancing at the paper and I don't want to do that right here so I want to read every sentence and understand it okay so here then I find out about the loss and then I if I don't know something here then I'll go and look it up on maybe on Wikipedia or something like this now I don't need to | 1,498 | 1,524 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1498s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | understand every single part of it right that's maybe I should correct myself so for example this bounding box loss here they talk about the second part of the max across the Hungarian Falls is this box loss that scores bounding boxes unlike many detectors that do box prediction with some Englishmen the other yada-yada they say the most commonly used l1 loss will have | 1,524 | 1,547 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1524s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | different scales for a small so here they basically talk about how they mix the losses they say overall our box loss is that defined as this and this now I haven't I don't know what these losses are I just assume there are some bounding box losses so when I it's not true when I say understand everything understand the things that are integral to the story of the paper right how | 1,547 | 1,572 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1547s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | exactly they compute bounding box losses at this point I don't care I just assume that there is some loss that I can back propagate right I what is important is that they do this Hungarian matching thing right as soon as I get that I'm like ah that was this you know this um this thing no this thing up here this thing this with the matching thing now I get it now I know uh there are always | 1,572 | 1,600 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1572s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | the same amount of boxes here there are always the same amount of labels here and all we need to do is somehow match them and I immediately think why is that relevant oh because when something is already matched to an object some other thing cannot be matched to the same object and that's how we you know prevent the fact that all the things predict the same thing right and so that immediately | 1,600 | 1,627 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1600s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | becomes clear and as I said there is usually like one or two ideas in a paper I don't assume or I don't care what their exact loss function is because I've sort of gotten the idea up here of what the loss is about alright so I hope that's clear under very closely read the things and understand the things that are necessary for the story if you find if you think something's not necessary | 1,627 | 1,655 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1627s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | for the story and then later end up not understanding that maybe come back and you know read it again in any case I would I would rather I would rather skip something and assume it's not necessary if I think so and then come back then trying to understand every everything but the things I do read I try to understand thoroughly okay then there's the architecture okay and | 1,655 | 1,684 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1655s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | that again I read closely and get backbone ok transformer encoder ok and now I understand much more closely a decoder ok and here I get now finally I get what this is about because n objects in parallel yada yada yada these input embeddings are learned positional encodings that we refer to as object queries and similarly to the encode we add them to the input at each attention | 1,684 | 1,713 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1684s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | layer so now they name i've already seen these object queries here and the only word i actually need from this sentence are learnt the fact that their positional encodings I just kind of ignore as soon as they say learnt I know AHA these things here are learned they have actually they're always the same for each of the images they're just over all learned okay so now I feel I | 1,713 | 1,738 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1713s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | understand the entire model and yeah so then they say auxiliary decoding losses and this sometimes you have to pay attention to like exhilarating because those are the things that here they say explicitly we found helpful to use auxiliary losses sometimes they they won't say why they did it they will just say our loss consists of three things and you know if you look at the three | 1,738 | 1,771 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1738s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | things only one of the things is really a part of their story so far and that you should immediately conclude that they've put in the other things because they tried it and it didn't work right so you can also kind of get an estimate of the brittleness and so on of the system in that you see how many unnecessary things are there or how many things are not straightforward how many | 1,771 | 1,794 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1771s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | things aren't the easiest thing that you would do when you would go about and do what they did okay so then you this concludes this model or method usually the student section is called like method or model or something like this and you go to experiments now the main question I have so far or I have maybe I have some more questions about the model itself that I haven't been able to pick | 1,794 | 1,820 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1794s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | up from this section which is not the case here but I simply keep those questions in mind and see whether they are resolved later right so I keep an awareness of what I don't understand but from here on my main issue is are they demonstrating that their story works right so they're here they're they're proposing a loss and a model and in my mind they now need to convince me that | 1,820 | 1,853 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1820s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | that works and that's that's it's not as easy as simply to show me some numbers that they are good at some benchmark they need to show me that they get those numbers because of what they claim so here they claim well okay they propose a new they propose a new architecture so what they need to convince me of is that the architecture itself makes sense right but in other papers when when you | 1,853 | 1,883 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1853s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | propose like and when you say like oh we for example in an L STM when you build in an attention mechanism and you claim oh we you know the attention mechanism can look back at the source sequence in one step then you need to convince me that that actually happens right so you need to not only need you need to perform well you need to convince me that you perform well because of what | 1,883 | 1,911 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1883s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | you claim your model does right so and that's often difficult and I specifically look out in the experiments for usually the question is like where are they trying to me right where are they trying to are or are they trying to me are they trying to cover up the fact that something doesn't work now all the experiments are always in the best light possible of course and | 1,911 | 1,938 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1911s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | you have to keep that in mind but a lot of times you can also already see from the experiments that okay are they doing something weird are they not showing me some obvious experiment or and that's a lot of time because is there an easier explanation for why they get the results that they get other than their explanation right and it is it is their job to convince you that their | 1,938 | 1,967 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1938s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | explanation is the correct one for these numbers and especially if there is an easier one that they haven't excluded and then I don't believe the experiments if that's the case right if there is an easier explanation for the effect I'm I'm very skeptical but some papers have an easier job here than other papers so in this paper they basically show results on a on a on a task and since | 1,967 | 1,995 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1967s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | their paper is about hey our pipeline is just easier than other pipelines what they first of all need to do is they need to like match the numbers of other pipelines and here I see that okay in these results often in the aftermath maybe a table or something here you see like this their model other models and their model is the best model in a lot of cases now if the best thing is of | 1,995 | 2,022 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=1995s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | course if is if their model throughout is the best the worst thing is if it's like scattered like this even if their model is the best but in every single benchmark a different configuration of their model is the best that's that's sort of a bad sign unless they can explicitly explain why that is and it's also not that good of a sign if these things are spread out like this like | 2,022 | 2,048 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2022s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | sometimes this baseline is good sometimes their model is better and so on so pay attention to that now in this paper it doesn't matter so much that's actually fine because what they're trying to show is that their model is on par and way easier and they've already made the case in what way it is easier it's easier in terms of architecture if there were to say it's much faster then | 2,048 | 2,072 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2048s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | after that I would expect you know an experiment in speed while these numbers are matched so but since they say it's easier I've already seen the architecture I'm convinced of that now that they show okay our numbers match actually I'm surprised they even outperform a lot of times then I'm quite happy with these experiments so also look for differences between numbers and | 2,072 | 2,098 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2072s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | the spread of numbers now it's not easy to say what if like point 1 is a bigger a small difference that depends on the task but if you know pay attention to these things pay attention to the fact that these results are noisy and oftentimes there is a lot more hyper parameter tuning going into the model of the paper then into the baseline model so why do you want to make your look | 2,098 | 2,122 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2098s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | your stuff look as good as possible and here is a little bit where the institutional credibility of someone like Facebook comes in in that I tend to believe their results a bit more than other results megha but a bit more yeah also look at patterns that they don't point out in the text so if there is like a pattern if you see like an interaction between the number of parameters and the score | 2,122 | 2,149 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2122s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | or something like this just try to be on the lookout of that and see if you can spot something that you think or think about whether that makes sense or not in what your hypothesis would be so here we go on and okay then they go into ablations and a lot of a lot of these papers do appellations and i generally appreciate that so here they visualize that the attention mechanism in their | 2,149 | 2,181 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2149s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | model actually refers to different instances right encoder self attention for a set of reference points the encoder is able to separate individual instances and you can see that pretty clearly right here where and even here with the overlapping cows and this is the sort of experiment that I would expect that actually convinces me that their architecture does what it says | 2,181 | 2,207 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2181s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | that it does right and something like this where you see like totally overlapping things with the attention of the individual things visualized so telling me like especially this one right here the the foot of the back elephant actually being focused by the attention of the bounding box of the back elephant that's the sort of experiment that convinces me that their | 2,207 | 2,232 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2207s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | claims like that their numbers really come from what they claim it comes from okay so at the end of the experimental section you should always ask yourself have they really convinced me that their story is true right that the improvement or winnig whenever they get an improvement or whatever they get whether is is due to the story that they want to sell me or could there be like an easier | 2,232 | 2,263 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2232s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | explanation or does something not fit is like other are the experiments different than from what you would SPECT here okay so these are these are my main questions are they are they convincing me of their story it's not do they have state of the art numbers I don't care I don't care even though like sometimes so there is a bit of a catch I I don't care about state of the art | 2,263 | 2,291 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2263s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | numbers now let's say you have a table like this and you have a computer vision model and one of the models is like on the C for 10 data set now if your baseline model has like a ninety one ninety two percent accuracy on C for ten when I know the state of the art is 96 I don't care right I know like I've done C for ten I know with like I don't know five six layer CNN you can reach these | 2,291 | 2,319 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2291s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | 91 92 93 % accuracy and to get to the 96 97 you would actually be like in the region of a wide ResNet and whatnot so I you know I know that even though you're a few points behind state of the art I know you know this this is valid still so I don't care but if you were to be like at 80 percent accuracy on C for 10 then I then I get a bit like I like it's pretty easy to get to 90 percent plus | 2,319 | 2,354 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2319s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | with like a standard CNN so there I immediately start to wonder why is there an explanation now this could be like a theoretical paper that says oh we investigate MLPs and that's why we only get that number so that's that would be fine but if something is out of the ordinary like this then I pay attention but never because something isn't like the latest and greatest state of the | 2,354 | 2,381 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2354s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | earth that's just dumb ok and also if only evaluate what the paper claims it does right if the paper says we want to show that we are on par with current models then don't be mad if the paper doesn't outperform these models they didn't claim that right so yeah so after these ablations I'm actually pretty happy right here with the results and this right here when I saw this I didn't | 2,381 | 2,413 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2381s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | I didn't expect that but I read the experiment description that these are these different learned object queries and what they do and that gave me an increased understanding of how these object queries actually work right so at that point I still had like a vague I knew that these are learned but reading this and sort of looking at it studying it a bit I was like oh okay then I | 2,413 | 2,438 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2413s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | understood even better what they are so again when I say understand everything in the method section you can still have questions and but you just have to keep it in mind for later and then here I go on and there's this de TR for Panoptix segmentation and they here they propose like a new model so I first look at it and I'm like okay they proposed a new model they can do stuff like this now | 2,438 | 2,467 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2438s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | this is not object detection and again I'm not sure is this like a is this like a an add-on to the method or is was was this up here just an intermediate step to this and honestly after reading that I still wasn't sure it seems like something in between of course the paper is also a bit longer than other papers it just it seems it's too long for just being a side note but it's too short for | 2,467 | 2,496 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2467s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | being its own thing so that was just a bit weird and I treated it as just like a oh we can also do this with our model but I didn't pay like too much attention to that okay so at the end I you know look at conclusions now the conclusions of a paper are much much often they are not nearly as informative as the introduction the conclusions they all often tend to be very generic and kind | 2,496 | 2,529 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2496s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | of hedging a bit against criticisms saying what would be up for future work which is again hedging against criticism because you could simply say well we didn't do this that's future work yes so again I read it but I don't really pay attention to it and then I gloss over the abstract I just would kind of scroll through the abstract if there's something that catches my eye I would | 2,529 | 2,556 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2529s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | look at it and if not then not and then I basically go to the start and whenever I didn't understand something I go back I look at it again and I try to think are all my questions answered and have they sufficiently convinced me that their story is the thing that really has the effect right here and then if I now were to make a video of this I've often found it useful to just put the paper | 2,556 | 2,588 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2556s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | away for a while and it's I usually get the best result when I read the paper the day before and then make a video the day after or if not I'll just you know put it away do something else do some email responding programming going outside eating lunch just some kind of a break between first read or between your first couple of reads and just I don't even think about the paper I just it's | 2,588 | 2,617 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2588s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | kind of it's in the subconscious it kind of bruised right and I happen to think about the paper every now and then but I don't make a conscious effort to be like oh how am I gonna explain this and so on but I just found the the worst videos are the ones where I immediately make the video after reading a paper and I've just discovered that if I kind of take a | 2,617 | 2,640 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2617s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | break and then I look at it again right I look I don't read it fully again but I if I have if I have the feeling I've understood it I don't read it fully again but I just kind of look at it and go again through this story and I think that's even if you you know wanna if you want to talk about a paper in a reading group or tell you know explain it to your friends or whatnot this is often | 2,640 | 2,662 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2640s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | very useful just put it away for a while let it Mellow and I find that helps a lot okay that was my process of reading this particular paper now we again this this is a high quality paper so it's I find it's a pretty easy read in that I simply need to understand what they did and I'm pretty happy with their experiments I maybe next time I can find an experiment | 2,662 | 2,691 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2662s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | or a paper where I'm initially more skeptical and not as happy with what I find but yeah let me know if you enjoy this or if you would like to see any other explanation I don't exactly know if this is what you expect it from a video like this so let me know maybe I have misunderstood you completely or it's way too long way too detailed or way too undetailed yeah leave me a | 2,691 | 2,720 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=2691s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
qHLLMg0Teg4 | all right welcome everyone welcome to lecture 13 of CS 287 last week on Thursday there was no live lecture but we actually did cover the material so ignosi did they recording at home and posted a video on Piazza so you can watch lecture 12 on video it will mean no live version of that lecture but this way we can keep up with the course and not lose a lecture slot today lecture 13 | 0 | 34 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=0s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | we'll look at Kalman smoothers maximum a-posteriori estimation maximum likelihood and expectation maximization all right so this is our menu for today let's start with smoothing what's the main idea behind smoothing let's go back to filtering which we've been doing in filtering what you have is you try to find a distribution over variable XT after you have observed some sensor | 34 | 62 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=34s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | measurements Z 0 through ZT and ideally over time I keep track of this when you go to the next time you again find a distribution for XT plus 1 given all observations up to time T plus 1 and repeat now if you look at this very symmetric you only use information from the past and often that's indeed the best you can do because right now you need to know the state of the robot as | 62 | 83 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=62s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | well as possible so the best you can do is filtering in that situation but what if you're post-processing your data you already collected your data I want to know where was my robot back at time T we should also use the information that happened after and so that's called smoothing and smoothing you use all the sensor observations before and after to come to a conclusion about the | 83 | 104 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=83s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | distribution over state of your robot or your environment now and the figures had true here I ignored the actions and in principle it could be action everywhere for every time slice but it does not change the math in any way what does the action do the only thing the action does if we're doing estimation is that the conditional distribution of XT plus 1 given X T becomes conditional XT plus 1 | 104 | 131 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=104s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | given XT and the action but once the action is fixed because you have the whole sequence is just indexing into a conditional distribution and putting it there index by the action and so let's assume that already happened we already have a conditional distribution or XT given XT minus 1 and maybe what's indexed by the action maybe there was no actions it doesn't really matter the | 131 | 150 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=131s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | math is the same all right so now to think about smoothing and compared to filtering let's work out the basics on the board so we'll do this by example rather than having a you know just very general like any kind of horizon problem we'll pick a very specific horizon just horizon 2 so we'll be interested in this up [Music] we'll be interested in the probability | 150 | 189 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=150s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | distribution over state at time two so that's our example in general would be any time T but we're going to do by example for time to given observations at time zero one and two now one thing we know is that this is proportional to the Joint Distribution over x2 and the observation Z 0 Z 1 Z 2 again what does this sign mean the proportional sign what it means is that we're looking at a | 189 | 226 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=189s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | distribution over x2 and what it means then we can evaluate this quantity here for all values of x2 and once we have all the numbers for every possible value of x2 we can just sum those together divide by that sum and that will normalize it make this sum to 1 and get the actual conditional now the model we have is this hmm like models so how do we write out this thing over here it's | 226 | 254 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=226s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | equal to the sum over variables we have ignored because we have a chain in how the probability distribution is set up and we left out x0 and x1 but as you write out the Joint Distribution they'll appear in it so we'll sum out over x0 and x1 and then we can write out the full chain rule for this hmm there is z2 given x2 before that there was x2 given x1 then with that there was | 254 | 290 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=254s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | right before that Z 1 given x1 and then before that there was x1 given x0 and before that there was z 0 given x0 and then distribution for X 0 and the graph corresponding to this is X 0 as to capitals x0 x1 x2 and then observed Z 0 C 1 Z 2 so we just throw it out the fool joint over all six variables here but then we only care about these four so we sum out over X 0 | 290 | 333 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=290s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | and X 1 to get distribution over just a 4 we care about now what do we do in filtering well what can I should start looking at some reorganization here X 0 where does it appear X 0 only appears at the end here so as we look at this summation we can actually move it we can say this is really X 0 summation can happen in the back here so sum over X 0 P X 1 given X 0 p 0 given X 0 P X 0 | 333 | 369 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=333s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | because as far as X 0 is concerned everything up front is a constant and bringing up a constant up front is fine to do that's just saying multiplying every term with a constant or first summing it all together and then multiplied with the constant is the same thing then how about X 1 X 1 does not appear in here so we can bring this up front so we can bring P Z 2 given X 2 up | 369 | 394 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=369s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | front and then we have the summation over X 1 and then this one does have X 1 in it and so that's this one and then this one after we sum out over X 0 we'll have X 1 and Z 0 in it C 0 is a constant that X 1 will be in it so we have to keep it in the back behind this summation now let's give these some meaning and how we would actually run the filtering algorithm | 394 | 419 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=394s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | we'd say okay this thing here is actually the joint is to join between Z 0 well let's write X first join between X 0 and Z 0 then we multiply with X 1 given X 0 and we sum out over X 0 we've seen this last week this will give us the joint between X 1 and Z 0 then we multiply with Z 1 given X 1 so multiply with this conditional at this point we have C 1 comma X 1 comma Z | 419 | 458 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=419s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | 0 then we multiply with X 2 given X 1 so here we'd have P X 2 comma Z 1 comma X 1 comma Z 0 then we sum out over X 1 so it disappears after here we have P X 2 comma Z 2 comma sorry C 1 comma C 1 C 0 and then here we multiply Z 2 given X 2 so at this point we have P Z 2 comma X 2 comma Z 1 comma Z 0 which is indeed what we have over here and so what we see is | 458 | 505 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=458s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | that we can recursively compute the Joint Distribution between the latest state variable X 2 and all past observations and the calculation we do is pretty much the same every time we just multiply in an observation they will multiply in the dynamics model and sum over the past variable then we multiply in the next observation multiplied and dynamics some out over | 505 | 530 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=505s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | the past state variable and repeat so we have a general recursive approach to finding these things and let's see we're going to put this let me put it over here we have P XT plus 1 comma Z 0 through ZT is equal to sum over X T the dynamics model XT plus 1 given X T times X T comma Z 0 through CT and then to bring in that last observation ZT plus 1 which is ultimate what we want here for | 530 | 578 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=530s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | that update now we have Z 0 through Z T n ZT plus is going to be multiplying this thing with the condition of observation given everything else so P Z T plus 1 given X T plus 1 times what we have here P X T plus 1 comma Z 0 through CT now these are bit equations are very easy to run and that's all you need to do to run filtering that's a quick reminder when we go from the joint so here we have the | 578 | 616 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=578s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | joint when XT plus 1 is e 0 through Z T and here we have a baton ZT plus 1 in general we need to multiply with the conditional of ZT plus 1 given all the variables already present here not just this one variable need to multiply with the new variable condition all variables you already have but in this model we have a conditional independence we don't need a condition on the past | 616 | 641 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=616s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | observations if we have the current state and so that's why we just have XT plus 1 here and not XT plus 1 and all observations to condition on because once we know XT plus 1 those past observations don't need to be conditioned on anymore they're independent of Z T plus 1 and that's shown in the graph structure here and we're looking at we're getting z2 given | 641 | 662 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=641s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | everything in the past is the same thing as z2 just given x2 and that's exactly what's happening here why we only condition on the XT plus 1 so that's filtering and we cover them but we covered it now in a different way that last I'm slightly a different way so we can now match up smoothing with what we saw here in filtering I'm gonna have to use a slightly smaller font for this | 662 | 687 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=662s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | example actually what a mm-hmm how much smaller brand is in your memories I'll use the entire width of the board as needed to cover smoothing so you've got this and we'll use the full width so smoothing now we care about let me try out here the context we will have x2 again and after the Y should be x3 and x4 and there will be observations Z for Z 3 Z 2 and I should all be things | 687 | 749 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=687s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | before it also x1 x0 and observation z1 c0 now we're just it in the distribution for x2 given all observations this is something you can do after the fact I have the fact analysis what happened to my system now to have all the information available so the quantity we're after is P x2 given Z 0 Z 1 Z 2 Z 3 Z 4 now again we know this is proportional to the joint between all | 749 | 794 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=749s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | these variables x2 comma Z 0 Z 1 Z 2 Z 3 Z 4 now we don't directly have this kind of joint available what we have available here is something of the form that involves all of these variables so all the conditionals multiplied together in that graph over there is what we have available and then some variables we're not going to care about X 0 X 1 X 3 and X 4 we don't care about we will write | 794 | 826 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=794s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | out the full joint over all 10 variables and then sum out the four variables we don't care about so what's the fool joint the fool joint would be Z 4 given X 4 that's happening all the way at the end then X for given X 3 then Z 3 given X 3 x3 given X to Z 2 given X 2 and it will continue on this other board X 2 given X 1 the C 1 given X 1 X 1 given X 0 z 0 given X 0 and then the prior P X 0 this | 826 | 888 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=826s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | is the full joint now for this full joint we know we don't care about X 0 X 1 X 3 and X 4 so we're going to sum them out to get rid of them X 0 X 1 X 3 X 4 we sum out over to get this quantity over here alright now we're going to play the same trick it's during when we sum out over these variables is their way to rearrange this into smaller calculations we're going to first | 888 | 916 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=888s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | multiply everything together and then finally get to some out but we can do smaller bite-sized calculations where we sum some variables out make them disappear and that way not have this exponential blow-up as the thing becomes bigger if you do it naively this way this summation that the variables are binary let's say then there will be two to the horizon number of terms in this | 916 | 936 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=916s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | summation but by bringing them in in the right spots you reduces and you get a linear competition in the length of the network rather than an exponential calculation so can we do the same thing here well let's see how about X 0 actually we can play the same trick as before X 0 instead of summing over it over here where does it appear only in the back here so let's just insert that | 936 | 961 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=936s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley | |
qHLLMg0Teg4 | summation over here how about X 1 all of these are constants as far as X 1 is concerned so we can put the summation over X 1 over here and get rid of it over here then as we think of this thing here so we have this calculation happening as a first calculation and then after that we can do this calculation which we know from filtering this is giving us the quantities we saw in filtering because | 961 | 1,000 | https://www.youtube.com/watch?v=qHLLMg0Teg4&t=961s | Lecture 13 Kalman Smoother, MAP, ML, EM -- CS287-FA19 Advanced Robotics at UC Berkeley |