Vision Transformer Quick Guide - Theory and Code in (almost) 15 min

Поділитися
Вставка
  • Опубліковано 17 гру 2024

КОМЕНТАРІ • 68

  • @JessSightler
    @JessSightler 7 місяців тому +8

    I've changed the output layer a bit... this:
    self.head_ln = nn.LayerNorm(emb_dim)
    self.head = nn.Sequential(nn.Linear(int((1 + self.height/self.patch_size * self.width/self.patch_size) * emb_dim), out_dim))
    Then in forward:
    x = x.view(x.shape[0], int((1 + self.height/self.patch_size * self.width/self.patch_size) * x.shape[-1]))
    out = self.head(x)
    The downside is that you'll likely get a lot more overfitting, but without it the network was not really training at all.

    • @DeepFindr
      @DeepFindr  7 місяців тому +5

      Hi, thanks for your recommendation.
      I would probably not use this model for real world data as there are many important details that are missing (for the sake of providing a simple overview).
      I will pin your comment for others that also want to use this implementation.
      Thank you!

  • @geekyprogrammer4831
    @geekyprogrammer4831 Рік тому +8

    This is very underrated channel. You deserve way more viewers!!

  • @betabias
    @betabias 6 місяців тому +1

    Keep making content like this, I am sure you will get a very good recognition in the future. Thanks for such amazing content.

  • @hmind9836
    @hmind9836 Рік тому +5

    You're awesome man!!! I clicked your video so fast, you're one of the my favorite AI youtubers. I work in the field and I think you have a wonderful ability of explaining complex concepts in your videos

    • @DeepFindr
      @DeepFindr  Рік тому

      thanks for the kind words :)

  • @florianhonicke5448
    @florianhonicke5448 Рік тому +4

    Really great explanation. Nice visuals

  • @tenma5220
    @tenma5220 Рік тому +1

    This channel is amazing. Please continue making videos!

  • @hemanthvemuluri9997
    @hemanthvemuluri9997 Рік тому

    Awesome man!! You code and explain with such simplicity.

  • @xxyyzz8464
    @xxyyzz8464 3 місяці тому +2

    Why use dropout with GeLU? Didn’t the GeLU paper specifically say one motivation for GeLU was to replace ReLU+dropout with a single GeLU layer?

  • @kristoferkrus
    @kristoferkrus Рік тому +4

    Nice video! However, I think it's incorrect that you would get separate vectors for the three channels? This is not how they do it in the paper; there they say that the number of patches is N = HW/P^2, where H, W and P is the height and width of the original image and (P, P) is the resolution of each patch, so the number of color channels doesn't affect the number of patches you get.

  • @vero811
    @vero811 7 місяців тому

    I think there is a confusion between cls token and positional embedding? At 6:09?

  • @netanelmad
    @netanelmad Рік тому

    Thank you! Very clear and informative.

  • @romanlyskov9785
    @romanlyskov9785 10 місяців тому

    Awesome! Thanks for excellent explanation!

  • @KacperPaszkowski-s4b
    @KacperPaszkowski-s4b Рік тому +3

    Hello, first of all great tutorial video. I've tried running provided code for training, but after ~400 epochs loss is still the same (~3.61) and model always predicts the same class. Do you have an idea what might be a problem with it?

    • @DeepFindr
      @DeepFindr  Рік тому

      Hi, have you tried a Lower learning rate? Also is the train loss decreasing or also stuck?

    • @KacperPaszkowski-s4b
      @KacperPaszkowski-s4b Рік тому +3

      ​@@DeepFindr Actually I've already found one bug in notebook. In forward method of Attention module, input is directly passed to MultiheadAttention bypassing linear layers.
      Changing learning rate doesn't affect training at all. Also, when training I've noticed that model's output converges to all zeros.
      I've checked gradients in network and it turns out that gradient flow stops at PatchEmbedding layer. All layers after it have non-zero gradients. Still don't know why this happens

    • @DeepFindr
      @DeepFindr  Рік тому

      Thanks for finding this bug. But I actually think it's not super relevant for this issue - I experimented with the attention previously and tried both ways (with linear layers and without), that's how this bug was created in the first place.
      When I started the training back then the loss was definitely decreasing, but I didn't expect it to get stuck at some plateau.
      Typically when models predict always the same class there can be a couple of reasons. I already checked this:
      - Input data is normalized
      - Too few / too many parameters (I would recommend to count the model parameters to get a feeling for this)
      - Learning rate
      - SGD Optimizer (seems to work a bit better)
      - Batch size, I put it to 128
      - Embedding size, make it a bit smaller
      After 100 epochs the Loss is also converging to 3.61, but the model is predicting different classes. Maybe the Dataset is not big enough. What about trying another dataset? Alternatively, try data augmentation.
      As stated in the video, transformers need to see a lot of examples.

  • @cosminpetrescu860
    @cosminpetrescu860 11 місяців тому +7

    Why are the positional embeddings learnable? It doesn't make sense to me

    • @trendingtech4youth989
      @trendingtech4youth989 11 місяців тому +1

      Bcoz, positional embedding represent the adress or position of image information of image patches

    • @cosminpetrescu860
      @cosminpetrescu860 11 місяців тому

      @@trendingtech4youth989 in "Attention is all you need", afaik, positioal embeddings are not learnable

    • @Omsip123
      @Omsip123 7 місяців тому

      ​@trendingtech4youth989 so..they are given like the patches, why shall they be learned

    • @sohangundoju8940
      @sohangundoju8940 7 місяців тому +4

      Imagine you have a sentence "I made a pizza and put it in an oven, it was tasty".
      Here we know that it refers to pizza, but for a model, it could mean pizza is delicious or oven is delicious.
      Positional encodings are learnt based on each other, the position of it here is with respect to pizza and oven. Therefore it is a learnt parameter.
      Here the video mentions that it is doesn't have any significant advantage over numbering, but when you are trying to teach the model to identify features, it will identify feature and position of one patch wrt other patches, ergo making it a learnable parameter

    • @isaakcarteraugustus1819
      @isaakcarteraugustus1819 6 місяців тому +2

      It’s not really needed, in the original transformer paper have been tests with learnable pe and basic-not-learnable-sinusoidal pe and that did not make a big difference, gpt2 also uses learnable pe, nowadays it’s more RoPE doing the positional encoding

  • @beratcokhavali
    @beratcokhavali Рік тому +1

    in 05:08. how we calculated that? when I calculated the patch shape I got a different result. Could someone explain that?

    • @chandank5266
      @chandank5266 7 місяців тому

      yes exactly, I also have same doubt, for me its 192 instead of 324

    • @user-wm8xr4bz3b
      @user-wm8xr4bz3b 6 місяців тому

      original image size was 144 (h) x 144 (w), after patch embedding, it was transformed to 324 patches, and each patch has 128 dimension size. 324 was derived from (144/8) x (144/8) = 18 x 18 = 324 patches. 8 is the patch size of each patch.

  • @dfdiasbr
    @dfdiasbr Місяць тому

    Awesome! Thanks for this video!

  • @comunedipadova1790
    @comunedipadova1790 День тому

    Has anyone been able to make it converge? What hyperparameters did you modify?

  • @anightattheraces
    @anightattheraces Рік тому

    Very helpful video, thanks!

  • @frommarkham424
    @frommarkham424 Місяць тому

    As a robot myself, i can confirm that an image really is worth 16x16 words

  • @gayanpathirage7675
    @gayanpathirage7675 6 місяців тому +1

    There was a error on your published code but not in the video.
    attn_output, attn_output_weights = self.att(x, x, x)
    It should be
    attn_output, attn_output_weights = self.att(q, k, v)
    Anyway, thanks for sharing the video and code base. It helped me a lot while learning ViT

  • @VoltVipin_VS
    @VoltVipin_VS 2 місяці тому

    The best part of Vision transformers is inbuilt support interpretability as compared to CNN where we had to compute saliency maps.

  • @adosar7261
    @adosar7261 5 місяців тому

    Isn't the embedding layer redundant? I mean we have then the projection matrices meaning that embedding + projection is a composition of two linear layers.

  • @frommarkham424
    @frommarkham424 Місяць тому

    An image is worth 16x16 words🗣🗣🗣🗣🗣🗣🗣💯💯💯💯💯💯💯🔥🔥🔥🔥🔥🔥🔥

  • @kitgary
    @kitgary Рік тому +1

    Awesome video! But I wonder if you reverse the order of LayerNorm and Multi-Head Attention? I think the LayerNorm should be applied after Multi-Head Attention but your implementation apply the LayerNorm before it.

    • @DeepFindr
      @DeepFindr  Рік тому +6

      Hi! Thanks!
      There is a paper that investigated pre- VS post-layernorm in transformers (see here arxiv.org/pdf/2002.04745). The "Pre" variant seems to perform better as opposed to the traditional suggestion in the transformer paper. This is also what most public implementations do :)

  • @josephmargaryan
    @josephmargaryan 6 місяців тому

    Is this better for the MNIST challenge compared to a simple conv network like LeNet

  • @datascienceworld
    @datascienceworld 10 місяців тому

    Great tutorial

  • @murphy1162
    @murphy1162 Рік тому

    Hope you could explain the swim transformer object detection in new video please

  • @justsomeone3375
    @justsomeone3375 Рік тому

    can someone help me with the training codes in the Google Colab link in the description?

  • @efexzium
    @efexzium 6 місяців тому

    can you please make a video on how to perform inference on VIT like googles open source vision transformer?

  • @MrMadmaggot
    @MrMadmaggot 9 місяців тому

    Is the Colab using cuda? IF so how can I tell if it is useing cuda

  • @abhinavvura4973
    @abhinavvura4973 7 місяців тому

    hi there
    I have used the code for binary class classification, but encountering the problem on accuracy , showing 100%
    accuracy only on label 1 and some times on label 2. So it would be helpful for me if u provide me any solution

    • @DeepFindr
      @DeepFindr  7 місяців тому

      Hi, please see pinned comment. Maybe this helps :)

  • @muhammadtariq7474
    @muhammadtariq7474 7 місяців тому

    Where to get slides? Used in video

  • @chinnum9716
    @chinnum9716 Рік тому

    Hey,great video. I have a question though. Isn't the entire point of 'pre' norm is that the normalization is applied before attention computation is performed?
    But from the code ,norm = PreNorm(128, Attention(dim=128, n_heads=4, dropout=0.)) , it seems like you are performing attention first and then normalizing aka post-norm. Please correct me if I'm wrong :)

    • @DeepFindr
      @DeepFindr  Рік тому +1

      Hi! In the forward pass of the PreNorm layer is this line:
      self.fn(self.norm(x), **kwargs)
      So normalization is applied first and then the function (such as attention in this example)
      The line you are referencing is just the initialization, not the actual call
      Hope that helps :)

    • @chinnum9716
      @chinnum9716 Рік тому

      @@DeepFindr stupid of me to not see that first. Thank you for the reply

  • @marcossrivas
    @marcossrivas Рік тому

    Cool video! What do you think about the implementation of ViT on signal processing (spectrogram analysis) , applied to audio for example. Which advantages could it have over the classic convolutional networks?

    • @sorvex9
      @sorvex9 7 місяців тому

      Take a look at gpt4-omni to find out, lol

  • @ycombinator765
    @ycombinator765 4 місяці тому

    bro is educated!

  • @newbie8051
    @newbie8051 4 місяці тому

    Ah, tough to understand, guess will have to read more on this to fully understand

    • @VoltVipin_VS
      @VoltVipin_VS 2 місяці тому

      You should have a deep understanding of transformer architecture to understand this.

  • @지능시스템트랙신현수
    @지능시스템트랙신현수 10 місяців тому

    감사합니다!!

  • @hautran-uc8gz
    @hautran-uc8gz 9 місяців тому

    thank you

  • @avirangal2044
    @avirangal2044 7 місяців тому

    The video is great but the training in the code didn't work for the entire 1000 epochs. Despite the code looks logical, there is endless of things that can go wrong so I think it was better to do a tutorial with working ViT notebook.

    • @DeepFindr
      @DeepFindr  7 місяців тому

      Hi! I think this is because the Dataset is too small. Transformers are data hungry. It should work with a bigger dataset

    • @DeepFindr
      @DeepFindr  7 місяців тому

      Also have a look at the pinned comment, maybe that helps :)

  • @RAZZKIRAN
    @RAZZKIRAN Рік тому

    thank u ,

  • @0x00official
    @0x00official 10 місяців тому +2

    And now sora uses the same algorithm. this video aged so well

    • @simpleplant606
      @simpleplant606 9 місяців тому +5

      Sora is using DiT (Diffusion Transformer)

  • @Saed7630
    @Saed7630 5 місяців тому

    Bravo!