title
stringlengths 12
112
| published
stringlengths 19
23
| url
stringlengths 28
28
| video_id
stringlengths 11
11
| channel_id
stringclasses 5
values | id
stringlengths 16
31
| text
stringlengths 0
596
| start
float64 0
37.8k
| end
float64 2.18
37.8k
|
|---|---|---|---|---|---|---|---|---|
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t954.64
|
especially for this so we have this image text data set and in here we don't have there's not
| 954.64 | 968.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t961.4399999999999
|
much it's just 21 images or text to image pairs and we can see what they look like so
| 961.44 | 975.12 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t969.1999999999999
|
we have this text aeroshock of a futuristic city with a large motorway okay so i tried to just
| 969.2 | 984.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t975.12
|
describe this image as as best i could and yeah that's what i got and there are like as like you
| 975.12 | 993.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t984.24
|
saw just now 21 of these image text pairs in there so let's go ahead and actually prepare or download
| 984.24 | 1,005.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t993.2
|
and sort of initialize clip for our for our use so the the model id on hooking face is this
| 993.2 | 1,015.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1005.6
|
so if we were to go to hooking face.co we could type that in here and we have the model there
| 1,005.6 | 1,022.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1015.04
|
okay so this is the model that we're using over from openai here and with this model we we use
| 1,015.04 | 1,031.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1022.48
|
these two we use a processor and a model so this is the model itself this is clip right this is a
| 1,022.48 | 1,039.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1031.04
|
almost like a pre-processor for both our text and also the images okay so one thing we would do here
| 1,031.04 | 1,046.08 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1039.92
|
if we have a CUDA device available we can move our model to the CUDA device
| 1,039.92 | 1,053.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1047.68
|
at the moment if you try and do this with nps so if you're on mac and you have a you have apple
| 1,047.68 | 1,059.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1053.04
|
silicon there are some processors or some transformations in the clip that don't
| 1,053.04 | 1,066.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1059.6000000000001
|
function on nps at the moment so i would stick with cpu we're only doing inference so it's still
| 1,059.6 | 1,073.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1066.72
|
pretty fast now as i was mentioning the the processor is what handles both the text and
| 1,066.72 | 1,080 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1073.84
|
image preparation that needs to happen before we feed them into the actual encoder models themselves
| 1,073.84 | 1,088 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1080.0
|
that make up clip so for text we do this so this is just going to be this is going to work like a
| 1,080 | 1,094.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1088.0
|
normal text tokenizer a normal text tokenizer for text transform models is used in order to
| 1,088 | 1,105.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1094.96
|
translate our human readable text into transformer readable ids okay so we pass the text here we make
| 1,094.96 | 1,110.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1105.8400000000001
|
sure we are saying there are no images included here because the processor if we have both
| 1,105.84 | 1,118.4 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1111.52
|
images and text it can process them at the same time we can do that here as well but i want to
| 1,111.52 | 1,122.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1118.4
|
show you it separately just to show you what they're actually doing so the padding
| 1,118.4 | 1,130.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1122.8
|
we need to set that to true and that is because different different sentences can have different
| 1,122.8 | 1,137.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1130.8
|
lengths okay so you have like hello world and whatever i wrote before up here so this
| 1,130.8 | 1,139.68 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1137.6
|
aerial shot of futuristic city
| 1,137.6 | 1,154.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1139.68
|
aerial shot of a city these two sentences have different lengths and a transform model needs to
| 1,139.68 | 1,161.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1154.24
|
see the same length being input for all of the the text that is within this sort of single batch
| 1,154.24 | 1,166.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1161.8400000000001
|
so basically what it's going to do there is add what are called padding labels so it's just going
| 1,161.84 | 1,176.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1166.96
|
to add a few of these up to the length of the longest sequence within that batch of of text
| 1,166.96 | 1,188.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1176.32
|
items because in here we have those 22 um 20 no sorry 21 sentences so that's all we're doing there
| 1,176.32 | 1,197.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1188.56
|
i'm sure that is uh and then we are returning those as pytorch sensors and then finally just
| 1,188.56 | 1,202.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1197.28
|
moving them to whichever device we're using i'm using cpu here so it's not actually necessary to
| 1,197.28 | 1,210.4 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1202.24
|
do this but i'm doing it in case you do do the same on a cuda enabled device so from there we
| 1,202.24 | 1,218.88 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1210.4
|
have these input ids and an attention mask okay so let's have a quick look at what what those are so
| 1,210.4 | 1,230.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1218.88
|
we go into tokens and we have a look at input ids okay you see we get all these literally just
| 1,218.88 | 1,240.16 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1230.8000000000002
|
integer values and you'll see that a lot of them have this 49407 at the end all right that is
| 1,230.8 | 1,245.12 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1240.16
|
they're the padding tokens there okay so they they are not represented as strings but they're
| 1,240.16 | 1,250.08 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1245.1200000000001
|
represented as these integer numbers okay and we know that they're the paying tokens because they
| 1,245.12 | 1,257.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1250.0800000000002
|
they're appearing several times at the end of each sequence and none of the sequences i fed in there
| 1,250.08 | 1,262.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1258.0
|
were they didn't have any similar words at the end of those okay so you can see them all here
| 1,258 | 1,268.08 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1262.72
|
so we know that those are the pattern sequences we also see there's like an initialization of
| 1,262.72 | 1,277.52 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1268.08
|
sequence token there as well and then everything in between those they are tokens that represent
| 1,268.08 | 1,284.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1277.52
|
a word or a part of a word from our original text so that's the input ids the attention mask
| 1,277.52 | 1,292.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1284.8
|
so that's the input ids the attention mask is you'll see so here you can see that it's just
| 1,284.8 | 1,300 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1292.72
|
these ones and zeros now the ones represent real tokens okay they represent real words that were in
| 1,292.72 | 1,310 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1300.0
|
our from our text inputs the zeros represent where the where our processor has added padding tokens
| 1,300 | 1,317.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1310.0
|
so this is used for the internal mechanisms of the text transformer to know which tokens to pay
| 1,310 | 1,322.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1317.2
|
attention to which ones to ignore because we don't want to really focus on those padding tokens because
| 1,317.2 | 1,329.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1322.48
|
they're meaningless they're just there to make sure we have the same size inputs going into our
| 1,322.48 | 1,339.12 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1329.2
|
transform model that's all that is so we can go down and after we have our tokens you know what
| 1,329.2 | 1,346.64 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1339.12
|
we what we do is we use clip to encode all of them with this get text features okay and then
| 1,339.12 | 1,352.08 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1346.6399999999999
|
we pass our tokens and i've got two device here i think i already i already moved them to device
| 1,346.64 | 1,360.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1352.08
|
so i don't need to do that again we can actually remove that okay and okay what do we get here
| 1,352.08 | 1,371.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1360.24
|
so we get 21 so 21 text inputs that makes sense 512 dimensional vectors okay so they are our text
| 1,360.24 | 1,378 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1371.6
|
embeddings representing each of those those text sentences that we just gave and then one other
| 1,371.6 | 1,384.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1378.0
|
thing i wanted to point out here is that we have the min and max values and they're pretty big okay
| 1,378 | 1,390.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1384.48
|
they're clearly not normalized so this depends on what you're doing if you are if you want to
| 1,384.48 | 1,398.88 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1390.96
|
compare these vectors you need to make sure you're not using a similarity metric that looks or that
| 1,390.96 | 1,406.16 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1398.88
|
considers the magnitude of your vectors you need to only consider the the angle so you can do that
| 1,398.88 | 1,411.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1406.16
|
with cosine similarity or the alternative is that you can normalize these vectors and then you can
| 1,406.16 | 1,419.68 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1411.36
|
also do this with dot product similarity okay so to normalize if you wanted to use that product
| 1,411.36 | 1,426.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1419.6799999999998
|
similarity now you would do this okay so here we're just detaching our text embeddings from the
| 1,419.68 | 1,431.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1426.4799999999998
|
the pytorch graph moving them cpu if needed we actually don't need to do that but do it here
| 1,426.48 | 1,438.16 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1431.84
|
anyway and convert them into a non-py array and then we calculate the value that we will normalize
| 1,431.84 | 1,443.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1438.16
|
that we will normalize it each vector by okay so for each each vector we're calculating a
| 1,438.16 | 1,452 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1444.88
|
number and then that number is what we're going to divide them all by here okay to to normalize that
| 1,444.88 | 1,462.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1452.72
|
and then after that you can see the minimum maximum is this minus 0.15 and plus 0.53 okay so
| 1,452.72 | 1,468 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1462.5600000000002
|
neither of them going over minus one or plus one now now when it comes to encoding images we we do
| 1,462.56 | 1,475.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1468.0
|
the same thing or very similar thing so images are also pre-processed using the using the processor
| 1,468 | 1,480.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1475.2
|
as we did with our text but we just use slightly different parameters to start there so the reason
| 1,475.2 | 1,487.52 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1480.8
|
we're processing these images is that clip expects a certain size of image when when we're feeding
| 1,480.8 | 1,496.64 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1488.08
|
images into it and expects those those image pixels to be normalized as well now rgb images
| 1,488.08 | 1,503.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1496.64
|
by default the the value the pixel values and they will range from zero to 255 we need to
| 1,496.64 | 1,508.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1503.6000000000001
|
normalize those and we also need to resize the images so you can see you can see that here so
| 1,503.6 | 1,514.88 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1508.96
|
the first image it has this size it's it's a pretty big image okay this is the the width and
| 1,508.96 | 1,520.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1514.88
|
the height of that image now here we're taking all the images and we're processing them make
| 1,514.88 | 1,526.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1520.8000000000002
|
sure we say text is is none and that will actually only output one tensor the pixel values tensor so
| 1,520.8 | 1,530.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1526.24
|
we're just going to extract that straight out there and we're also going to move it to the
| 1,526.24 | 1,536.16 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1530.96
|
device set hardware device in this case just cpu and now let's have a look at this image
| 1,530.96 | 1,543.76 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1536.16
|
or images now so now we can see that we have this this array or tensor with three color channels so
| 1,536.16 | 1,552 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1543.76
|
this is the rgb and it has a height and width of 224 so it's been you know sort of squeezed into
| 1,543.76 | 1,561.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1552.0
|
a smaller size now and we have 21 days because we fed in all of our images okay so this is how we
| 1,552 | 1,567.52 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1561.04
|
use the processor and this is just resizing and normalizing our images ready for the division
| 1,561.04 | 1,575.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1567.52
|
transformer encoder of clip and very similar to before before we use get text features now we're
| 1,567.52 | 1,582.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1575.44
|
going to use get image features and we pass in those images like that and again as you you might
| 1,575.44 | 1,592.4 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1582.3200000000002
|
expect those images are not normalized you see that here and as we would also expect they are
| 1,582.32 | 1,598 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1592.4
|
the same dimensionality as our text embeddings so that means we can compare them but before
| 1,592.4 | 1,603.76 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1598.64
|
comparing them of course as before we we normalize them so we should normalize them again here
| 1,598.64 | 1,612.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1603.76
|
um and yep same process again and we can see that those have those have changed okay cool so
| 1,603.76 | 1,621.76 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1613.6
|
what we now want to do is calculate the similarity between all of our image embeddings and all of our
| 1,613.6 | 1,629.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1621.76
|
text embeddings so we can do that in a few different ways we have cosine similarity or
| 1,621.76 | 1,633.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1629.2
|
dot product similarity the reason we can use our product similarity is because we normalize
| 1,629.2 | 1,637.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1633.8400000000001
|
but i'm going to show you how to do both so that if you don't normalize you can actually just use
| 1,633.84 | 1,644.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1637.92
|
a cosine similarity like we do here so cosine similarity is actually just a dot product as
| 1,637.92 | 1,650.88 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1644.32
|
a numerator between the text embeddings and image embeddings and in the denominator we have just
| 1,644.32 | 1,660.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1650.88
|
normalized the norm values of both of those okay that is that's all it is actually so it's it's
| 1,650.88 | 1,666.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1660.3200000000002
|
pretty pretty simple and if we plot those similarity scores between those we get this
| 1,660.32 | 1,672.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1666.24
|
so we would expect along this diagonal here we'd expect these to be the highest similarity values
| 1,666.24 | 1,678.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1672.72
|
which say represent the the true pairs okay between the images and the text now we have some
| 1,672.72 | 1,684.4 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t1678.56
|
that are not quite there like here and there is this image text pair which is more similar even
| 1,678.56 | 1,690.48 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.