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-t275.28000000000003
|
makes sense that that is the sort of direction that machine learning and AI may also go in.
| 275.28 | 290.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t282.8
|
So to achieve this multi-modality in CLIP we actually use two models that are trained to
| 282.8 | 296.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t291.52000000000004
|
almost speak the same language. So with these two models one of them is a text encoder one of them
| 291.52 | 304.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t296.72
|
is an image encoder. Both of these models create a vector representation of whatever they are being
| 296.72 | 310 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t304.32000000000005
|
input so the text encoder may get a sentence that sentence could be two dogs running across a frosty
| 304.32 | 320.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t310.0
|
field and then we have a image of two dogs running across a frosty field and CLIP will be trained so
| 310 | 329.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t320.8
|
that the text encoder consumes our sentence and outputs a vector representation that is very very
| 320.8 | 337.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t329.6
|
closely aligned to what the image encoder has output based on the image of the same concept.
| 329.6 | 344.88 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t337.44
|
Now by training both of these models to encode these vectors into a similar vector space we
| 337.44 | 351.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t344.88
|
are teaching them to speak the same vector language right so this is it's very abstract
| 344.88 | 359.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t351.04
|
this this vector language is like 512 dimensional space so we can't directly understand what
| 351.04 | 362.64 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t359.28
|
or it's very difficult for us to directly understand what is actually happening there
| 359.28 | 373.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t363.68
|
but these two models do actually output patterns that are logical and and make sense and we can
| 363.68 | 378.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t373.36
|
see some of this by comparing the similarity between the vectors that it outputs okay so we
| 373.36 | 384.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t378.96000000000004
|
can see that the two vectors for dogs running across a frosty field both the the text vector
| 378.96 | 392.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t384.8
|
and the image vector are both within a very similar vector space whereas something else like elephants
| 384.8 | 399.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t392.32
|
in the Serengeti is you know whether it's text or image is not here with our our two dogs running
| 392.32 | 404.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t399.28
|
across the frosty field is somewhere over over here right in a completely different space so
| 399.28 | 409.52 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t404.47999999999996
|
what we can do with that is is calculate the similarity between these vectors and identify
| 404.48 | 416.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t409.52
|
which ones are similar or not similar according to clip from this from these these meaningful
| 409.52 | 421.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t416.32
|
vectors that that clip is actually outputting we are able to create a content-based image retrieval
| 416.32 | 428.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t421.91999999999996
|
system okay so content-based image retrieval is basically where we um using some text or using
| 421.92 | 438 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t428.72
|
maybe even another image we can search for images based on their content right and not just like
| 428.72 | 443.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t438.0
|
some metatextual metadata or something that's been attached to it and with clip unlike other
| 438 | 450.88 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t443.20000000000005
|
content-based image retrieval systems um clip is incredibly good at actually capturing the meaning
| 443.2 | 457.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t450.88000000000005
|
across the entire image so you know for example with our our two dogs running across a frosty field
| 450.88 | 461.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t457.04
|
we might also be able to describe the background of that image without mentioning that there's two
| 457.04 | 467.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t461.44
|
dogs in it and if we describe in such a way that um we align pretty well with what that image
| 461.44 | 472.96 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t467.36
|
actually is what is in that image we might actually also return the image based on that so
| 467.36 | 479.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t472.96000000000004
|
we're not just focusing on one thing in the image clip allows us to focus on many things in the image
| 472.96 | 487.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t479.44
|
so an example of that is within this data set i've been using here there are no images there's
| 479.44 | 495.52 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t487.92
|
one single image of the food a hot dog okay so i tried to search that and the first image that is
| 487.92 | 501.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t495.52
|
returned is a dog eating a hot dog okay so it's pretty relevant but of course there are no other
| 495.52 | 506.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t501.36
|
images of hot dogs in this in this data set so the other images that are returned are quite
| 501.36 | 513.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t506.72
|
interesting because in some way or another they are kind of showing a hot dog so the first one we
| 506.72 | 521.76 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t513.6
|
have a dog looking pretty cozy in a warm room with a fire in the background then we have a dog in a
| 513.6 | 529.68 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t521.76
|
big woolly jumper and another dog kind of like posing for the camera so weirdly enough we we
| 521.76 | 535.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t529.6800000000001
|
got a load of hot dog images even though it's not really um maybe it's not exactly what we meant when
| 529.68 | 541.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t535.2
|
we said hot dog but a person could understand that okay we can we can see how those that term
| 535.2 | 547.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t541.6
|
and those images are related now we're not actually only restricted to text to image search
| 541.6 | 552.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t547.6
|
when we encode our our data when we code text and when we code images we are actually just
| 547.6 | 559.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t552.8000000000001
|
creating vectors so we can search across that space in any any direction with any combination
| 552.8 | 566.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t559.92
|
of modalities so we could do a text to text search image to image search we can also do image
| 559.92 | 571.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t566.24
|
to text search or we can search everything we could use some text to search for text and images
| 566.24 | 577.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t572.3199999999999
|
we can kind of go in any direction use any modality we want now let's go into a little
| 572.32 | 585.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t577.8399999999999
|
more detail on what the architecture of clip actually looks like so clip as i mentioned it's
| 577.84 | 591.6 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t585.84
|
these two models now these two models are trained in parallel one of them is the the text encoder
| 585.84 | 598.08 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t591.6
|
now it's a just a generic text encoder of 12 layers and then on the image encoder side there
| 591.6 | 604.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t598.08
|
are there are two different options i've spoken about there is a vision transformer model and
| 598.08 | 611.76 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t604.72
|
also a resnet model and they use a few different sizes for resnet as well both of these both of
| 604.72 | 619.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t611.76
|
these encoder models output a single 512 dimensional vector and the way these models
| 611.76 | 625.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t619.2
|
is trained is is kind of in the name of clip so clip stands for contrastive learning in pre-training
| 619.2 | 632.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t626.0
|
and so the the training that is used during pre-training is is contrastive it's contrastive
| 626 | 639.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t632.56
|
pre-training now across both nlp and computer vision large models sort of dominate the the
| 632.56 | 645.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t639.36
|
state of the art and the reason for this or the idea behind this is that just by giving a large
| 639.36 | 652.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t645.28
|
model a huge amount of data they can learn sort of general patterns from what they see and almost
| 645.28 | 661.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t652.5600000000001
|
kind of internalize a a general rule set for the the data that it sees okay so they manage to
| 652.56 | 668.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t662.48
|
recognize general patterns in their modality in language they may be able to internalize the
| 662.48 | 677.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t668.72
|
grammar rules and patterns in english language for vision models that may be sort of the general
| 668.72 | 682.72 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t678.5600000000001
|
patterns that you identify or notice in with different scenes and different objects
| 678.56 | 689.12 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t683.28
|
now the problem with these different models the reason they don't fit together very well already
| 683.28 | 696.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t689.76
|
is that they're trained separately so by default these state of the art models have no understanding
| 689.76 | 701.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t696.24
|
of each other and that that's where clip is is different that's what clip has has brought to the
| 696.24 | 710.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t701.84
|
table here with clip the text and image encoders are trained while considering the context of the
| 701.84 | 717.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t710.48
|
other modality okay so the text encoder is trained and it considers the modality or it considers the
| 710.48 | 722.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t717.2
|
the concept learned by the image encoder and the image encoder does the same for the text encoder
| 717.2 | 729.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t722.56
|
and we can almost think of this as the the image and text encoders are sharing a almost indirect
| 722.56 | 737.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t729.4399999999999
|
understanding of the other modality now contrastive training works by taking a image and
| 729.44 | 743.12 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t737.1999999999999
|
text pair so for example the two dogs running across a frosted field and putting those together
| 737.2 | 749.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t743.1199999999999
|
into the text encoder and image encoder and learning to encode them both as as closely as
| 743.12 | 759.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t749.92
|
possible for this to work well we also need negative pairs so we need something to compare
| 749.92 | 764.8 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t759.4399999999999
|
against this is a general rule in contrastive learning you can't just have positive pairs
| 759.44 | 769.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t764.8
|
because then everything can just be kind of encoded into the same like tiny little space
| 764.8 | 776.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t769.8399999999999
|
and you don't know how to separate the the pairs are dissimilar okay so we need both positive and
| 769.84 | 782.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t776.24
|
negative pairs so we have a positive pair okay in order to get negative pairs we can essentially
| 776.24 | 792.16 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t782.5600000000001
|
just take all the positive pairs in our data set and we can say okay the pair t1 and i1 we can mix
| 782.56 | 800.56 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t792.16
|
t1 with different eyes okay so we can do t1 with i2 and i3 i4 and so on so we're basically just
| 792.16 | 807.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t800.56
|
swapping the pairs and we can we can understand that there's other pairs are probably not going
| 800.56 | 813.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t807.3599999999999
|
to be similar as long as our data set is relatively large occasionally maybe we will get a pair that
| 807.36 | 818.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t813.4399999999999
|
are similar but as long as our data set is large enough that that doesn't happen too frequently
| 813.44 | 825.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t818.3199999999999
|
it's not going to affect our training it will be sort of a negligible problem so with this idea
| 818.32 | 834.64 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t825.92
|
we can use a loss function that will minimize the difference between positive pairs and maximize the
| 825.92 | 840.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t834.64
|
difference between negative pairs and that will look something like this where we have our positive
| 834.64 | 847.04 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t840.3199999999999
|
pairs in the diagonal of the similarity matrix and everything else is something that we the dot
| 840.32 | 854.48 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t847.04
|
product there we need to maximize and this image that you see here is actually the pre-training for
| 847.04 | 861.36 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t854.48
|
a single batch okay so one interesting thing to note here is if we have a small batch say we only
| 854.48 | 866.32 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t861.36
|
have a batch size of two it's going to be very easy for our model to identify which two items are
| 861.36 | 873.92 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t866.32
|
similar which two are not similar whereas if we have 64 in our 64 items in our batch it will be
| 866.32 | 879.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t873.9200000000001
|
much harder for our model because it has to it has to find more nuanced differences between them and
| 873.92 | 887.52 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t879.84
|
and what basically the odds of guessing randomly between those and guessing correctly are much
| 879.84 | 895.68 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t887.52
|
smaller so a larger batch size is a good thing to to aim for in this contrastive pre-training
| 887.52 | 904.08 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t895.6800000000001
|
approach so with that i think we we have a good idea now of how clip can be used and also you
| 895.68 | 911.28 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t904.08
|
know how it has been trained for for this so what i really want to do now is kind of show you how you
| 904.08 | 917.44 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t911.2800000000001
|
might be able to use it as well now we're going to be using the vision transformer version of clip
| 911.28 | 923.84 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t918.0
|
okay so we remember i said there's a the resnet and vision transformer options for that image encoder
| 918 | 931.68 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t923.84
|
we're going to use a vision transformer version and openai have released this model through the
| 923.84 | 936.24 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t931.68
|
hooking face library so we can we can go to the hooking face library and use it directly from now
| 931.68 | 941.2 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t936.2399999999999
|
which makes it really easy for us to actually sort of get started with it so let's go ahead and do
| 936.24 | 948 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t941.1999999999999
|
that now okay so for this we will need to install a few libraries here so we have transformers
| 941.2 | 954.64 |
CLIP Explained | Multi-modal ML
|
2022-09-15 13:00:22 UTC
|
https://youtu.be/fGwH2YoQkDM
|
fGwH2YoQkDM
|
UCv83tO5cePwHMt1952IVVHw
|
fGwH2YoQkDM-t948.64
|
torch and data sets so data sets we need to actually get data set so i've prepared one
| 948.64 | 961.44 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.